{ "id": 613699, "title": "Welcome to the CSIRO - Image2Biomass Prediction Competition!", "url": "/competitions/csiro-biomass/discussion/613699", "author_name": "Qiyu Liao", "post_date": "2025-10-29T02:24:53.221000", "votes": 31, "comment_count": 117, "views": 0, "content": "

We are thrilled to host this challenge because we believe the brightest data minds from around the globe can help solve one of the most persistent and meaningful problems in modern agriculture.

\n

If you have any questions about the data, evaluation criteria, or any other aspect of the competition, please feel free to post them in the Discussion forum. We welcome all questions and discussions, and I will do my best to answer them promptly.

\n

Thank you for joining us, and we wish you the very best of luck!

", "messages": [ { "id": 3308275, "postDate": "2025-10-29T02:24:53.220Z", "content": "

We are thrilled to host this challenge because we believe the brightest data minds from around the globe can help solve one of the most persistent and meaningful problems in modern agriculture.

\n

If you have any questions about the data, evaluation criteria, or any other aspect of the competition, please feel free to post them in the Discussion forum. We welcome all questions and discussions, and I will do my best to answer them promptly.

\n

Thank you for joining us, and we wish you the very best of luck!

", "rawMarkdown": "We are thrilled to host this challenge because we believe the brightest data minds from around the globe can help solve one of the most persistent and meaningful problems in modern agriculture.\n\nIf you have any questions about the data, evaluation criteria, or any other aspect of the competition, please feel free to post them in the Discussion forum. We welcome all questions and discussions, and I will do my best to answer them promptly.\n\nThank you for joining us, and we wish you the very best of luck!", "votes": 31 }, { "id": 3381378, "postDate": "2025-12-24T11:50:20.677Z", "content": "

Thank you for organizing this interesting competition. I would like to share some feedback and suggestions regarding the prize pool distribution. Currently, the prizes are listed as:

\n
1st Place – $50,000  \n2nd Place – $20,000  \n3rd Place – $5,000  \nTotal: $75,000\n
\n

Based on the current leaderboard and related discussions, it appears that significant rank fluctuations are likely. In such a scenario, the current prize distribution may feel unfair to participants finishing in 4th or 5th place, especially if their scores are very close to those of the top three. It is also possible that the performance difference between the 1st- and 2nd-place models may not be substantial enough to justify such a large gap in rewards.

\n

I understand that this structure follows common Kaggle practices; however, given the competitive nature and tight score margins in this competition, it may warrant reconsideration. I would like to propose the following alternatives:

\n

Suggestion 1: Distribute the total prize pool among the top 5 or top 10 leaderboard positions. This approach would be both fairer and more aligned with the competitive dynamics of the leaderboard, while remaining Kaggle-friendly. For example:

\n
1st Place – $20,000  \n2nd Place – $15,000  \n3rd Place – $10,000  \n4th Place - ..\n
\n

Suggestion 2: After the competition concludes, use the final leaderboard scores of the top 5 or top 10 teams to guide a more nuanced prize distribution. For example, if the top scores are very close (e.g., 70%, 65%, and 64%), the rewards for adjacent ranks, such as 2nd and 3rd place should not differ significantly.

\n

I believe either approach would better reflect the actual performance differences and encourage broader participation.

", "rawMarkdown": "Thank you for organizing this interesting competition. I would like to share some feedback and suggestions regarding the prize pool distribution. Currently, the prizes are listed as:\n\n```\n1st Place – $50,000 \n2nd Place – $20,000 \n3rd Place – $5,000 \nTotal: $75,000\n```\n\nBased on the current leaderboard and related discussions, it appears that significant rank fluctuations are likely. In such a scenario, the current prize distribution may feel unfair to participants finishing in `4th` or `5th` place, especially if their scores are very close to those of the top three. It is also possible that the performance difference between the 1st- and 2nd-place models may not be substantial enough to justify such a large gap in rewards.\n\nI understand that this structure follows common Kaggle practices; however, given the competitive nature and tight score margins in this competition, it may warrant reconsideration. I would like to propose the following alternatives:\n\n**Suggestion 1:** Distribute the total prize pool among the top `5` or top `10` leaderboard positions. This approach would be both fairer and more aligned with the competitive dynamics of the leaderboard, while remaining Kaggle-friendly. For example:\n\n```\n1st Place – $20,000 \n2nd Place – $15,000 \n3rd Place – $10,000 \n4th Place - ..\n```\n\n**Suggestion 2:** After the competition concludes, use the final leaderboard scores of the top `5` or top `10` teams to guide a more nuanced prize distribution. For example, if the top scores are very close (e.g., 70%, 65%, and 64%), the rewards for adjacent ranks, such as 2nd and 3rd place should not differ significantly.\n\nI believe either approach would better reflect the actual performance differences and encourage broader participation.", "votes": 3 }, { "id": 3370030, "postDate": "2025-12-10T14:29:57.190Z", "content": "

Hi @qiyuliaocsiro, Can we use DINOv3 pretrained models? Its license is not OSI-approved and not a GPL3 license, but Section 2.5.a mentions exceptions for pretrained models with incompatible licenses. Please clarify if this applies here.

\n

License reference: https://github.com/facebookresearch/dinov3/blob/main/LICENSE.md

", "rawMarkdown": "Hi @qiyuliaocsiro, Can we use DINOv3 pretrained models? Its license is not OSI-approved and not a GPL3 license, but Section 2.5.a mentions exceptions for pretrained models with incompatible licenses. Please clarify if this applies here.\n\nLicense reference: https://github.com/facebookresearch/dinov3/blob/main/LICENSE.md", "votes": 3, "replies": [ { "id": 3370575, "postDate": "2025-12-10T23:58:42.047Z", "content": "

Hi Shivam, please refer to this topic: https://www.kaggle.com/competitions/csiro-biomass/discussion/617603#3322386

", "rawMarkdown": "Hi Shivam, please refer to this topic: https://www.kaggle.com/competitions/csiro-biomass/discussion/617603#3322386", "replies": [ { "id": 3374536, "postDate": "2025-12-12T18:18:13.657Z", "content": "

Hey @qiyuliaocsiro,

\n
    \n
  1. Is the weighted r2 computed as per Kaggle description (uses global ground truth mean and weighting in denominator) or the attached literature (computes R2 for each biomass component separately). Both result in different results.
    2. \n
  2. Are you evaluating our result on log-transformed values?
    3. \n
", "rawMarkdown": "Hey @qiyuliaocsiro, \n1. Is the weighted r2 computed as per Kaggle description (uses global ground truth mean and weighting in denominator) or the attached literature (computes R2 for each biomass component separately). Both result in different results. \n2. Are you evaluating our result on log-transformed values?" }, { "id": 3377199, "postDate": "2025-12-15T23:34:58.833Z", "content": "

Hi Naitik, Everything is subject to the Kaggle description page.

", "rawMarkdown": "Hi Naitik, Everything is subject to the Kaggle description page." } ] } ] }, { "id": 3387310, "postDate": "2026-01-06T17:54:16.560Z", "content": "

Hi @qiyuliaocsiro,

\n

I have a question regarding the use of test data during the inference phase.

\n

Since this is a Code Competition, are we permitted to use unsupervised techniques on the test set during the notebook runtime? Specifically, I am interested in:

\n

Transductive / Continual Learning: Updating model weights or batch normalization statistics dynamically using the test set features (without labels) during the inference run.

\n

Post-processing: adjusting final predictions based on the statistical distribution of the test data batch.

\n

I want to ensure my approach remains within the competition rules regarding data leakage.

", "rawMarkdown": "Hi @qiyuliaocsiro,\n\nI have a question regarding the use of test data during the inference phase.\n\nSince this is a Code Competition, are we permitted to use unsupervised techniques on the test set during the notebook runtime? Specifically, I am interested in:\n\nTransductive / Continual Learning: Updating model weights or batch normalization statistics dynamically using the test set features (without labels) during the inference run.\n\nPost-processing: adjusting final predictions based on the statistical distribution of the test data batch.\n\nI want to ensure my approach remains within the competition rules regarding data leakage.", "votes": 1, "replies": [ { "id": 3387839, "postDate": "2026-01-07T17:32:09.027Z", "content": "

Not an official answer but working on that exact thing today :) When I have tried it in the past Compute Time has always been an issue, in fact first attempt last night ran over the time limit and died this morning.

", "rawMarkdown": "Not an official answer but working on that exact thing today :) When I have tried it in the past Compute Time has always been an issue, in fact first attempt last night ran over the time limit and died this morning. " } ] }, { "id": 3364771, "postDate": "2025-12-06T17:44:54.717Z", "content": "

Qiyu

\n

Its not clear to me how the evaluation handled weeds in the study.

\n

For example, Crumbweed is bad - too much can make animals sick. One would assume that you do not want it listed as Biomass - but…

\n

Were known weeds not included in the grass or clover or dead values?

", "rawMarkdown": "[Qiyu](https://www.kaggle.com/qiyuliaocsiro)\n\nIts not clear to me how the evaluation handled weeds in the study. \n\nFor example, Crumbweed is bad - too much can make animals sick. One would assume that you do not want it listed as Biomass - but...\n\nWere known weeds not included in the grass or clover or dead values?", "votes": 1, "replies": [ { "id": 3370573, "postDate": "2025-12-10T23:56:32.213Z", "content": "

Hi Jimmy, that's another critical research topic. I think we'll have some following work for that.

\n

For now, we only consider the distinction between green grass and dead clover. Finer classifications, such as separating out specific weeds like crumbweed, are currently not feasible.

", "rawMarkdown": "Hi Jimmy, that's another critical research topic. I think we'll have some following work for that.\n\nFor now, we only consider the distinction between green grass and dead clover. Finer classifications, such as separating out specific weeds like crumbweed, are currently not feasible." } ] }, { "id": 3359489, "postDate": "2025-12-01T18:16:32.843Z", "content": "

I think the current dataset is excellent for one specific task:\ngiven an image + NDVI + height + date + species, predict the biomass at that moment.\nThat’s a well-defined ML problem, and many models will perform strongly on it.

\n

However, real grazing and pasture management are dynamic and path-dependent systems. Two paddocks with the same present biomass can behave very differently depending on their recent grazing pressure. Rainfall history, stress cycles, soil type, paddock identity, and past management all influence recovery and biomass regrowth has real hysteresis. Overgrazed paddocks do not recover like rested ones.

\n

None of this structure is represented in the dataset. There are no paddock IDs, no longitudinal sequences, no stocking density or grazing events, no soil properties, and no management history ,only static snapshots treated as independent samples. As a result, NDVI + height + seasonal timing already capture most of the explainable signal, and image texture provides some marginal improvement, but true pasture recovery dynamics are nearly impossible to learn from this format.

\n

For a future version, it would be incredibly valuable to include repeated measurements from the same paddock over time, explicit grazing/rest intervals, rainfall summaries, soil type, and stocking history. That would allow models to answer the questions farmers genuinely care about, not just “How much biomass is here today?” but “How will this paddock respond if I graze it now versus five days later?”

\n

Thanks to the organizers !!

", "rawMarkdown": "I think the current dataset is excellent for one specific task:\ngiven an image + NDVI + height + date + species, predict the biomass at that moment.\nThat’s a well-defined ML problem, and many models will perform strongly on it.\n\nHowever, real grazing and pasture management are dynamic and path-dependent systems. Two paddocks with the same present biomass can behave very differently depending on their recent grazing pressure. Rainfall history, stress cycles, soil type, paddock identity, and past management all influence recovery and biomass regrowth has real hysteresis. Overgrazed paddocks do not recover like rested ones.\n\nNone of this structure is represented in the dataset. There are no paddock IDs, no longitudinal sequences, no stocking density or grazing events, no soil properties, and no management history ,only static snapshots treated as independent samples. As a result, NDVI + height + seasonal timing already capture most of the explainable signal, and image texture provides some marginal improvement, but true pasture recovery dynamics are nearly impossible to learn from this format.\n\nFor a future version, it would be incredibly valuable to include repeated measurements from the same paddock over time, explicit grazing/rest intervals, rainfall summaries, soil type, and stocking history. That would allow models to answer the questions farmers genuinely care about, not just “How much biomass is here today?” but “How will this paddock respond if I graze it now versus five days later?”\n\nThanks to the organizers !!", "votes": 1, "replies": [ { "id": 3360212, "postDate": "2025-12-02T01:18:26.543Z", "content": "

I believe building a comprehensive multimodal model to directly address the practical questions farmers face and support reliable decision-making is our ultimate goal.

\n

As you can see, we are still some distance from achieving that in this research aera: real-world data is fragmented, difficult to align and integrate, and certain fundamental data dimensions are challenging to collect. Computer vision–based biomass estimation is an important step toward that vision—it enables large-scale, low-cost, and pottentially accurate biomass measurement. Only with such capability can we begin to meaningfully connect biomass with other factors and make robust management decisions.

\n

Thank you for your thoughtful comment. Your suggestions are highly valuable, and we will certainly consider them in our future work.

", "rawMarkdown": "I believe building a comprehensive multimodal model to directly address the practical questions farmers face and support reliable decision-making is our ultimate goal.\n\nAs you can see, we are still some distance from achieving that in this research aera: real-world data is fragmented, difficult to align and integrate, and certain fundamental data dimensions are challenging to collect. Computer vision–based biomass estimation is an important step toward that vision—it enables large-scale, low-cost, and pottentially accurate biomass measurement. Only with such capability can we begin to meaningfully connect biomass with other factors and make robust management decisions.\n\nThank you for your thoughtful comment. Your suggestions are highly valuable, and we will certainly consider them in our future work.", "votes": 1 }, { "id": 3369917, "postDate": "2025-12-10T12:57:08.993Z", "content": "

I thought it was to give an image only -> then predict NDVI + height + date + species + target

", "rawMarkdown": "I thought it was to give an image only -> then predict NDVI + height + date + species + target" } ] }, { "id": 3334673, "postDate": "2025-11-17T15:53:20.933Z", "content": "

In an attempt to understand the data I have tried to model clover and grass growth over the year. My models don't make sense! Do you know if the pastures used in the data collection were used for grazing during 2015?

\n

That would seem to be the most likely cause for my issues.

", "rawMarkdown": "In an attempt to understand the data I have tried to model clover and grass growth over the year. My models don't make sense! Do you know if the pastures used in the data collection were used for grazing during 2015?\n\nThat would seem to be the most likely cause for my issues.", "votes": 1, "replies": [ { "id": 3334801, "postDate": "2025-11-17T16:34:54.040Z", "content": "

I also just confused, why they include year column.

", "rawMarkdown": "I also just confused, why they include year column. " }, { "id": 3335547, "postDate": "2025-11-18T04:51:43.937Z", "content": "

I believe the grazing status of each sampling site can no longer be traced back and is, in fact, difficult to quantify.

", "rawMarkdown": "I believe the grazing status of each sampling site can no longer be traced back and is, in fact, difficult to quantify." } ] }, { "id": 3313558, "postDate": "2025-11-09T13:14:33.047Z", "content": "

Hi @qiyuliaocsiro ,\nCan we pretrain a model locally on additional datasets that are publicly available online?\nWould using these locally pretrained model weights for the official submission be allowed, or would it count as using external data against the competition rules?

", "rawMarkdown": "Hi @qiyuliaocsiro ,\nCan we pretrain a model locally on additional datasets that are **publicly available** online?\nWould using these locally **pretrained model weights** for the official submission be allowed, or would it count as using external data against the competition rules?", "votes": 1, "replies": [ { "id": 3313856, "postDate": "2025-11-09T23:13:03.470Z", "content": "

Yes, you are allowed to use any public datasets that are also available to other competitors, as stated in Rule 2.6.

", "rawMarkdown": "Yes, you are allowed to use any public datasets that are also available to other competitors, as stated in Rule 2.6." } ] }, { "id": 3308587, "postDate": "2025-10-29T17:50:08.673Z", "content": "

NDVI data's format? Per-image value or spatial map?

", "rawMarkdown": "NDVI data's format? Per-image value or spatial map?", "votes": 1, "replies": [ { "id": 3308661, "postDate": "2025-10-29T23:25:12.983Z", "content": "

Per-image value

", "rawMarkdown": "Per-image value" } ] }, { "id": 3308325, "postDate": "2025-10-29T05:27:34.420Z", "content": "

i am guessing this will have massive participation

", "rawMarkdown": "i am guessing this will have massive participation", "votes": 1 }, { "id": 3386151, "postDate": "2026-01-04T17:32:39.133Z", "content": "

Hello, could anyone please help me check why the code runs successfully, but the submission fails.

\n

\"\"

", "rawMarkdown": "Hello, could anyone please help me check why the code runs successfully, but the submission fails.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F27667739%2Fe74eaa772258028ca1eff97e22c11234%2FScreenshot%202026-01-04%20at%205.28.04pm.png?generation=1767547700190890&alt=media)", "votes": -1, "replies": [ { "id": 3386292, "postDate": "2026-01-05T01:43:53.037Z", "content": "

Put a link to your notebook after you make it public - makes it easier for folks to help. Without the notebook - suggest you read all the posts - use 'submission' in the search and you will see a bunch.

", "rawMarkdown": "Put a link to your notebook after you make it public - makes it easier for folks to help. Without the notebook - suggest you read all the posts - use 'submission' in the search and you will see a bunch.", "replies": [ { "id": 3387332, "postDate": "2026-01-06T18:36:18.603Z", "content": "

here is the link: https://www.kaggle.com/code/ammaruk/csiro-biomass-error-fix?scriptVersionId=288706820

\n

Thanks for helping

", "rawMarkdown": "here is the link: https://www.kaggle.com/code/ammaruk/csiro-biomass-error-fix?scriptVersionId=288706820\n\nThanks for helping" }, { "id": 3387350, "postDate": "2026-01-06T19:17:30.243Z", "content": "

The link is broken. I looked and don't see any public shared notebook by you.

", "rawMarkdown": "The link is broken. I looked and don't see any public shared notebook by you. " }, { "id": 3390513, "postDate": "2026-01-13T10:10:49.320Z", "content": "

I have made the book public, sorry I didn't know that.

\n

https://www.kaggle.com/code/ammaruk/csiro-biomass-error-fix/edit/run/288706820

", "rawMarkdown": "I have made the book public, sorry I didn't know that.\n\nhttps://www.kaggle.com/code/ammaruk/csiro-biomass-error-fix/edit/run/288706820" }, { "id": 3390546, "postDate": "2026-01-13T11:06:42.993Z", "content": "

Ok - can see your notebook now. But I cannot fork and run the notebook as your data sets are private. You will need to also make those public.

\n

It does seem that your doing the embedding on local machine and reading those from a file. You cannot embed the test images on local machine. As I read the code it does seem like your only going to do a single image.

\n

Have your tried predicting on the train data

\n
train_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/test.csv')\n
\n

change to

\n
train_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\n
\n

Common mistake is that the code will only handle the single test image - predicting on the train data exposes those errors.

", "rawMarkdown": "Ok - can see your notebook now. But I cannot fork and run the notebook as your data sets are private. You will need to also make those public.\n\nIt does seem that your doing the embedding on local machine and reading those from a file. You cannot embed the test images on local machine. As I read the code it does seem like your only going to do a single image.\n\nHave your tried predicting on the train data\n\n```python\ntrain_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/test.csv')\n```\n\nchange to\n\n```python\ntrain_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\n```\n\nCommon mistake is that the code will only handle the single test image - predicting on the train data exposes those errors.\n\n" }, { "id": 3390563, "postDate": "2026-01-13T11:48:31.133Z", "content": "

Actually I embedded the train images on Kaggle but then I got submission error and made a new notebook in which I removed the train_img_embedding (coding bit) because I downloaded the embeddings and uploaded them from local computer. Anyways, I have made the datasets public.

\n

Btw, if I am not putting the train_img_embedding code bit, can it effect my submission?

\n

Thanks for pointing it out: test_data = pd.read_csv('/kaggle/input/csiro-biomass/test.csv')

\n

I will change it.

", "rawMarkdown": "Actually I embedded the train images on Kaggle but then I got submission error and made a new notebook in which I removed the train_img_embedding (coding bit) because I downloaded the embeddings and uploaded them from local computer. Anyways, I have made the datasets public.\n\nBtw, if I am not putting the train_img_embedding code bit, can it effect my submission?\n\nThanks for pointing it out: test_data = pd.read_csv('/kaggle/input/csiro-biomass/test.csv')\n\nI will change it." }, { "id": 3390625, "postDate": "2026-01-13T14:48:28.173Z", "content": "

Guess I do not understand your test embeddings - assume you did that on local machine. Are they the result of the train images or did they come from the single test image we see?

\n

Forking and running your code now attempting to predict on train rather than test.

", "rawMarkdown": "Guess I do not understand your test embeddings - assume you did that on local machine. Are they the result of the train images or did they come from the single test image we see?\n\nForking and running your code now attempting to predict on train rather than test." }, { "id": 3390628, "postDate": "2026-01-13T14:50:21.443Z", "content": "

So - if you replace test.csv with train.csv and run the code in the editor you quickly get to your error. Your only generating a single image worth in the dataframe. This generates a shape issue a cell or two later.

\n
train_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\n
", "rawMarkdown": "So - if you replace test.csv with train.csv and run the code in the editor you quickly get to your error. Your only generating a single image worth in the dataframe. This generates a shape issue a cell or two later.\n\n```python\ntrain_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\n```" } ] }, { "id": 3396977, "postDate": "2026-01-26T08:27:14.600Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 3381440, "postDate": "2025-12-24T15:47:45.540Z", "content": "

Image2Biomass Prediction

", "rawMarkdown": "Image2Biomass Prediction", "votes": -10 }, { "id": 3399136, "postDate": "2026-01-30T09:05:01.703Z", "content": "

Hello @qiyuliaocsiro

\n

Why have my ranking and score been cancelled? I did not violate any rules during this competition, and I open-sourced my code in the early stages. Could the organizers please explain why I can no longer find my ranking? Thank you very much.

", "rawMarkdown": "Hello @qiyuliaocsiro\n\nWhy have my ranking and score been cancelled? I did not violate any rules during this competition, and I open-sourced my code in the early stages. Could the organizers please explain why I can no longer find my ranking? Thank you very much." }, { "id": 3399015, "postDate": "2026-01-30T04:02:35.833Z", "content": "

Why have my competition results disappeared, and why does it show that I did not participate in this competition?@addisonhoward @qiyuliaocsiro

", "rawMarkdown": "Why have my competition results disappeared, and why does it show that I did not participate in this competition?@addisonhoward @qiyuliaocsiro" }, { "id": 3397128, "postDate": "2026-01-26T14:44:23.377Z", "content": "

“”You've exceeded the number of hours allowed to use GPU. Your quota will reset to 30 hours in 4 days. You can continue working in this session until you've reached your 9 hour limit. You will not be able to commit unless you turn off GPU.“” how could i submit with gpu？

", "rawMarkdown": "“”You've exceeded the number of hours allowed to use GPU. Your quota will reset to 30 hours in 4 days. You can continue working in this session until you've reached your 9 hour limit. You will not be able to commit unless you turn off GPU.“” how could i submit with gpu？" }, { "id": 3392937, "postDate": "2026-01-17T21:16:06.110Z", "content": "

Hi, I have some questions. The train.csv file contains the values of NVDI, height and others, associated to each image of the dataset. However, the test.csv does not have those values. I mean, during scoring, will these values be provided? Should we avoid to build the model using those values?

", "rawMarkdown": "Hi, I have some questions. The train.csv file contains the values of NVDI, height and others, associated to each image of the dataset. However, the test.csv does not have those values. I mean, during scoring, will these values be provided? Should we avoid to build the model using those values?", "replies": [ { "id": 3392967, "postDate": "2026-01-17T23:54:20.777Z", "content": "

They will not be provided. If you want to use them you will need to create them, use means/medians or not use them.

", "rawMarkdown": "They will not be provided. If you want to use them you will need to create them, use means/medians or not use them." } ] }, { "id": 3392419, "postDate": "2026-01-16T23:38:48.190Z", "content": "

What's wrong with submission file get keep get this error : Notebook Threw Exception · 21m ago \nthis the link to y code : https://www.kaggle.com/code/mohameddrabo/csiro-image2biomass-prediction

", "rawMarkdown": "What's wrong with submission file get keep get this error : Notebook Threw Exception · 21m ago \nthis the link to y code : https://www.kaggle.com/code/mohameddrabo/csiro-image2biomass-prediction", "replies": [ { "id": 3392423, "postDate": "2026-01-16T23:49:43Z", "content": "

Lots of discussion posts on this error. I even have one. This will not work

\n
sample  = pd.read_csv('/kaggle/input/csiro-biomass/sample_submission.csv')\n
\n

Because kaggle screwed up (IMO) this file is not available in the private test folder when scoring submission. Your code will likely work if you use test.csv instead.

", "rawMarkdown": "Lots of discussion posts on this error. I even have [one](https://www.kaggle.com/competitions/csiro-biomass/discussion/665381). This will not work\n\n```python\nsample = pd.read_csv('/kaggle/input/csiro-biomass/sample_submission.csv')\n\n```\nBecause kaggle screwed up (IMO) this file is not available in the private test folder when scoring submission. Your code will likely work if you use test.csv instead." } ] }, { "id": 3391634, "postDate": "2026-01-15T10:47:27.207Z", "content": "

How should I use other features?

", "rawMarkdown": "How should I use other features?" }, { "id": 3390384, "postDate": "2026-01-13T04:02:57.150Z", "content": "

i met this problem“Couldn't verify your identity Persona couldn't verify your identity. If you feel this is in error, contact support.” all my submission disappeared，and i got unranked，but i got 243th. can you help me solve this issue?

", "rawMarkdown": "i met this problem“Couldn't verify your identity Persona couldn't verify your identity. If you feel this is in error, contact support.” all my submission disappeared，and i got unranked，but i got 243th. can you help me solve this issue?" }, { "id": 3390382, "postDate": "2026-01-13T04:00:56.520Z", "content": "

Hello, I would like to inquire about an issue with my account. My competition rankings and submission records have disappeared. Furthermore, the rankings from my past competitions have all changed to 'unrank'. Could you please look into this and resolve the issue?

", "rawMarkdown": "Hello, I would like to inquire about an issue with my account. My competition rankings and submission records have disappeared. Furthermore, the rankings from my past competitions have all changed to 'unrank'. Could you please look into this and resolve the issue?", "replies": [ { "id": 3390383, "postDate": "2026-01-13T04:02:24.927Z", "content": "

Probably not the best place to ask for this help.

", "rawMarkdown": "Probably not the best place to ask for this help. " } ] }, { "id": 3381180, "postDate": "2025-12-24T01:06:20.677Z", "content": "

Hi! I'm not sure I understand the evaluation criteria. Is it R2 of the entire train+test set using only images, or R2 of the test set using only images (contains 1 image)?

", "rawMarkdown": "Hi! I'm not sure I understand the evaluation criteria. Is it R2 of the entire train+test set using only images, or R2 of the test set using only images (contains 1 image)? ", "replies": [ { "id": 3381186, "postDate": "2025-12-24T01:33:08.577Z", "content": "

The test set that will be evaluated is unseen to us - right now only a single dummy image is in test - when you submit for a score it will be replaced by around 800 images. The train images are not evaluated for the LB score.

", "rawMarkdown": "The test set that will be evaluated is unseen to us - right now only a single dummy image is in test - when you submit for a score it will be replaced by around 800 images. The train images are not evaluated for the LB score." } ] }, { "id": 3378520, "postDate": "2025-12-18T12:22:06.773Z", "content": "

Hello, could you please help me check why the code runs successfully, but the submission fails and no score is obtained?<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5396927%2Fb8326621c34235dc04963051d5d8e63b%2F2.png?generation=1766060523804972&alt=media\" alt=\"! (opens in a new tab)\">\" />

", "rawMarkdown": "Hello, could you please help me check why the code runs successfully, but the submission fails and no score is obtained?![![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5396927%2Fb8326621c34235dc04963051d5d8e63b%2F2.png?generation=1766060523804972&alt=media)](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5396927%2F063337bde5ffd3e65b665a773e7bab7e%2F1.png?generation=1766060514514656&alt=media)", "replies": [ { "id": 3378521, "postDate": "2025-12-18T12:22:51.327Z", "content": "

\"\"

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5396927%2Fe53e7b8ecef7325356c93cb9a18e2834%2F2.png?generation=1766060569060952&alt=media)", "replies": [ { "id": 3378964, "postDate": "2025-12-18T22:54:28.030Z", "content": "

Couple of common reasons in this competition - 1) You cannot read sample_submission.csv as for some reason it does not exist in the private test folder 2) you have a shape issue that does not show up with single image in the public test.

\n

change to read the train.csv instead of test.csv so that more images are predicted. You will almost certainly see an error issue you can act on - fix that and than go back to reading test.

\n

That does not do the trick - make your notebook public and share a link here. Someone might spot the issue.

", "rawMarkdown": "Couple of common reasons in this competition - 1) You cannot read sample_submission.csv as for some reason it does not exist in the private test folder 2) you have a shape issue that does not show up with single image in the public test.\n\nchange to read the train.csv instead of test.csv so that more images are predicted. You will almost certainly see an error issue you can act on - fix that and than go back to reading test.\n\nThat does not do the trick - make your notebook public and share a link here. Someone might spot the issue." } ] } ] }, { "id": 3378490, "postDate": "2025-12-18T10:54:20.697Z", "content": "

@qiyuliaocsiro Since the test set is split by time, I was wondering if the time distribution between the Public and Private Leaderboards is uniform (i.e., do they both cover all four seasons)? Additionally, is the species distribution balanced between the two? I would be extremely grateful for your reply!

", "rawMarkdown": "@qiyuliaocsiro Since the test set is split by time, I was wondering if the time distribution between the Public and Private Leaderboards is uniform (i.e., do they both cover all four seasons)? Additionally, is the species distribution balanced between the two? I would be extremely grateful for your reply!" }, { "id": 3375145, "postDate": "2025-12-12T21:04:25.683Z", "content": "

Hi @qiyuliaocsiro, I am getting a Notebook Threw Exception error. Can you suggest me any ideas or solutions to debug it. Also, do you have any ideas what type of error it could be in Notebook Threw Exception. Thanks!🙏

", "rawMarkdown": "Hi @qiyuliaocsiro, I am getting a Notebook Threw Exception error. Can you suggest me any ideas or solutions to debug it. Also, do you have any ideas what type of error it could be in Notebook Threw Exception. Thanks!🙏 ", "replies": [ { "id": 3377200, "postDate": "2025-12-15T23:37:18.887Z", "content": "

Sorry, I'm not able to resolve any operational issues. Please contact the Kaggle official for support

", "rawMarkdown": "Sorry, I'm not able to resolve any operational issues. Please contact the Kaggle official for support" } ] }, { "id": 3371266, "postDate": "2025-12-11T11:36:17.870Z", "content": "

Hello,

\n

Grass Weight Prediction

\n

What is this?

\n

Predict how much grass weighs from photos of pastures.

\n

Input

\n\n

Output

\n

Predict 5 weights (in grams):

\n\n

Problem

\n

Test data only has photos, no extra info. Need to use images to predict.

\n

I do not understand. if the input are only images and that is it?

", "rawMarkdown": "Hello, \n# Grass Weight Prediction\n\n## What is this?\nPredict how much grass weighs from photos of pastures.\n\n## Input\n- Photos of grass fields\n- Some extra info: greenness score (NDVI), grass height, location, grass type, date\n\n## Output\nPredict 5 weights (in grams):\n- Total dry grass\n- Green grass\n- Dead grass \n- Clover\n- Fresh grass (GDM)\n\n## Problem\nTest data only has photos, no extra info. Need to use images to predict.\n\nI do not understand. if the input are only images and that is it?", "replies": [ { "id": 3371694, "postDate": "2025-12-11T16:00:54.337Z", "content": "

You can model only the images and predict. Or you can model the extra info and the images for predictions.

", "rawMarkdown": "You can model only the images and predict. Or you can model the extra info and the images for predictions. " }, { "id": 3385943, "postDate": "2026-01-04T09:41:45.617Z", "content": "

So the task basically becomes predicting target based on the image?\nWe can not use the additional feature while training as those information are not available for test dataset.

", "rawMarkdown": "So the task basically becomes predicting target based on the image?\nWe can not use the additional feature while training as those information are not available for test dataset.\n", "replies": [ { "id": 3386089, "postDate": "2026-01-04T15:34:45.037Z", "content": "

you could train a smaller model to predict those additional features and then include that as input to the final model

", "rawMarkdown": "you could train a smaller model to predict those additional features and then include that as input to the final model\n" } ] } ] }, { "id": 3348128, "postDate": "2025-11-25T16:14:15.600Z", "content": "

Hello, I would like to know if the test set has the same metadata as the training set?

", "rawMarkdown": "Hello, I would like to know if the test set has the same metadata as the training set?", "replies": [ { "id": 3348215, "postDate": "2025-11-25T17:05:52.127Z", "content": "

No - there is no metadata for the test set!

", "rawMarkdown": "No - there is no metadata for the test set!", "replies": [ { "id": 3349132, "postDate": "2025-11-26T13:16:10.927Z", "content": "

Thank you. I understand.

", "rawMarkdown": "Thank you. I understand." } ] } ] }, { "id": 3344863, "postDate": "2025-11-23T03:14:25.280Z", "content": "

today ,I don't know what's going on . users from china couldn't get on kaggle through chinese internet without vpn

", "rawMarkdown": "today ,I don't know what's going on . users from china couldn't get on kaggle through chinese internet without vpn" }, { "id": 3340613, "postDate": "2025-11-19T16:23:39.250Z", "content": "

Dear Qiyu Laio, I am having difficultie in submitting the file. I trun off internet of the setting panel, and it seem to be another tag on the right panel; so when knitting the submit button, i am having a message that I should turn off internet; so please which one should turn?

\n

Thank you

", "rawMarkdown": "Dear Qiyu Laio, I am having difficultie in submitting the file. I trun off internet of the setting panel, and it seem to be another tag on the right panel; so when knitting the submit button, i am having a message that I should turn off internet; so please which one should turn?\n\nThank you", "replies": [ { "id": 3341109, "postDate": "2025-11-20T00:03:01.127Z", "content": "

Sorry Sowhe, I can hardly help solve any submission or operation issues.\n@martynaplomecka Hi Martyna, can you help with this?

", "rawMarkdown": "Sorry Sowhe, I can hardly help solve any submission or operation issues.\n@martynaplomecka Hi Martyna, can you help with this?", "replies": [ { "id": 3342106, "postDate": "2025-11-20T16:31:58.763Z", "content": "

The internet must be turned off during submission. Please make sure it is disabled when submitting

", "rawMarkdown": "The internet must be turned off during submission. Please make sure it is disabled when submitting" }, { "id": 3362432, "postDate": "2025-12-04T17:35:11.207Z", "rawMarkdown": "", "isDeleted": true } ] } ] }, { "id": 3336617, "postDate": "2025-11-18T14:54:12.050Z", "content": "

Would it be better to display more digits after the decimal point in the evaluation metric score?

", "rawMarkdown": "Would it be better to display more digits after the decimal point in the evaluation metric score?" }, { "id": 3331053, "postDate": "2025-11-16T11:14:47.727Z", "content": "

Not able to submit notebook for prediction. Also there is no error to debug in logs. \nnotebook link: https://www.kaggle.com/code/vasant10/csiro-image2biomass-multi-output-regression

", "rawMarkdown": "Not able to submit notebook for prediction. Also there is no error to debug in logs. \nnotebook link: https://www.kaggle.com/code/vasant10/csiro-image2biomass-multi-output-regression", "replies": [ { "id": 3347437, "postDate": "2025-11-25T06:35:47.577Z", "content": "

I am having the same issue.

", "rawMarkdown": "I am having the same issue." }, { "id": 3362493, "postDate": "2025-12-04T19:03:01.620Z", "content": "

Vasabt

\n

Mehmet

\n

When submissions fails but running against the small test is ok - you should ALWAYS replace test with train and see if any memory, size assumptions, timing issues exist, etc. I changed the cell for reading the test file to this:

\n

test_df = pd.read_csv(\"/kaggle/input/csiro-biomass/train.csv\")

\n

This will generate an error for you to debug. After the code will predict on train than you can change it back to test.csv and try for a submission.

\n

Good luck.

", "rawMarkdown": "[Vasabt](https://www.kaggle.com/vasant10)\n\n\n[Mehmet](https://www.kaggle.com/kabamehmetali)\n\nWhen submissions fails but running against the small test is ok - you should ALWAYS replace test with train and see if any memory, size assumptions, timing issues exist, etc. I changed the cell for reading the test file to this:\n\ntest_df = pd.read_csv(\"/kaggle/input/csiro-biomass/train.csv\")\n\nThis will generate an error for you to debug. After the code will predict on train than you can change it back to test.csv and try for a submission.\n\nGood luck.\n\n" } ] }, { "id": 3320938, "postDate": "2025-11-13T00:05:16.700Z", "content": "

Does the train/test images have the same resolution?

", "rawMarkdown": "Does the train/test images have the same resolution?", "replies": [ { "id": 3320943, "postDate": "2025-11-13T00:13:44.467Z", "content": "

Yes, they are in the same resolution.

", "rawMarkdown": "Yes, they are in the same resolution.", "votes": 1 } ] }, { "id": 3311506, "postDate": "2025-11-05T06:42:54.447Z", "content": "

i am having a issue with the submission , when i am running the notebook it is generating model weights and submission.csv in the same output folder but when i am submitting the notebook , i am getting the response that notebook is running without interrupting but \nsubmission.csv is not there

", "rawMarkdown": "i am having a issue with the submission , when i am running the notebook it is generating model weights and submission.csv in the same output folder but when i am submitting the notebook , i am getting the response that notebook is running without interrupting but \nsubmission.csv is not there ", "replies": [ { "id": 3376905, "postDate": "2025-12-15T11:45:51.623Z", "content": "

same problem being faced here too during submission @qiyuliaocsiro

", "rawMarkdown": "same problem being faced here too during submission @qiyuliaocsiro ", "replies": [ { "id": 3377202, "postDate": "2025-12-15T23:38:43.953Z", "content": "

Sorry, I'm not able to resolve any operational issues. Please get in touch with the Kaggle official for support

", "rawMarkdown": "Sorry, I'm not able to resolve any operational issues. Please get in touch with the Kaggle official for support" } ] } ] }, { "id": 3311447, "postDate": "2025-11-05T03:27:31.223Z", "content": "

@qiyuliaocsiro please correct me.

\n

Dry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g,
\nGDM_g = Dry_Total_g - Dry_Dead_g

\n

so we have to predict Dry_Green_g , Dry_Clover_g , Dry_Dead_g.

\n

\"Dry_Dead_g\" is the biomass(clover + other_plants) which are completely dead and dried.

\n

\"Dry_Green_g\" is the biomass(other_plants) which are dried but alive.

\n

\"Dry_Clover_g\" is the biomass(clover) which are dried but alive.

", "rawMarkdown": "@qiyuliaocsiro please correct me.\n\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g, \nGDM_g = Dry_Total_g - Dry_Dead_g\n\nso we have to predict Dry_Green_g , Dry_Clover_g , Dry_Dead_g.\n\n\"Dry_Dead_g\" is the biomass(clover + other_plants) which are completely dead and dried.\n\n\"Dry_Green_g\" is the biomass(other_plants) which are dried but alive.\n\n\"Dry_Clover_g\" is the biomass(clover) which are dried but alive.", "replies": [ { "id": 3311854, "postDate": "2025-11-05T23:43:44.350Z", "content": "

Technically, yes. The Dry_Total_g and GDM_g in the annotations are indeed the sum of the other components.

\n

However, please note that our evaluation metrics are based on initial validation: separately predicting Dry_Green_g, Dry_Clover_g, and Dry_Dead_g, then summing them (calculated prediction), is likely less accurate than directly predicting the total amount (model prediction). On the other hand, the prediction of the total amount is the most critical for the task (as reflected in the evaluation metrics), which is why we have set five targets instead of three.

", "rawMarkdown": "Technically, yes. The Dry_Total_g and GDM_g in the annotations are indeed the sum of the other components.\n\nHowever, please note that our evaluation metrics are based on initial validation: separately predicting Dry_Green_g, Dry_Clover_g, and Dry_Dead_g, then summing them (calculated prediction), is likely less accurate than directly predicting the total amount (model prediction). On the other hand, the prediction of the total amount is the most critical for the task (as reflected in the evaluation metrics), which is why we have set five targets instead of three.", "votes": 3, "replies": [ { "id": 3360108, "postDate": "2025-12-01T21:31:21.963Z", "content": "

Also based on weights we have to do that but wanted to know the rationale if the math is right why wasnt the metric just based on the 3 dry labels instead of adding the total and GDM g (with ofcourse there individual weights) .

", "rawMarkdown": "Also based on weights we have to do that but wanted to know the rationale if the math is right why wasnt the metric just based on the 3 dry labels instead of adding the total and GDM g (with ofcourse there individual weights) . " } ] } ] }, { "id": 3311340, "postDate": "2025-11-04T19:10:18.323Z", "content": "

This is my first competition, i am having difficulty uploading my file. It passes but it says submission failed. anyone know what i could be missing?

", "rawMarkdown": "This is my first competition, i am having difficulty uploading my file. It passes but it says submission failed. anyone know what i could be missing?", "replies": [ { "id": 3311443, "postDate": "2025-11-05T03:07:00.240Z", "content": "

\"In this competition, we will privately re-run your selected Notebook Version with a hidden test set substituted into the competition dataset. We then extract your chosen Output File from the re-run and use that to determine your score.\" this is the hidden test based competition, did you check?

", "rawMarkdown": "\"In this competition, we will privately re-run your selected Notebook Version with a hidden test set substituted into the competition dataset. We then extract your chosen Output File from the re-run and use that to determine your score.\" this is the hidden test based competition, did you check?", "replies": [ { "id": 3362439, "postDate": "2025-12-04T17:38:28.190Z", "rawMarkdown": "", "isDeleted": true } ] } ] }, { "id": 3311044, "postDate": "2025-11-04T05:16:22.230Z", "content": "

hello, can we use metadata like NDVI or other fileds during training??

", "rawMarkdown": "hello, can we use metadata like NDVI or other fileds during training??", "replies": [ { "id": 3311046, "postDate": "2025-11-04T05:25:37.970Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 3313245, "postDate": "2025-11-09T02:00:40.370Z", "content": "

hey Aic could you please guide me to submit i am facing errors i have 73% accuracy

", "rawMarkdown": "hey Aic could you please guide me to submit i am facing errors i have 73% accuracy ", "isDeleted": true }, { "id": 3313335, "postDate": "2025-11-09T05:36:55.143Z", "rawMarkdown": "", "isDeleted": true } ] } ] }, { "id": 3310894, "postDate": "2025-11-03T20:36:33.763Z", "content": "

Do I need to submit a notebook along with the prediction.csv? The 'Submit Prediction' button does not work. Is there an alternative way of submitting predictions? Patiently waiting for a response.

", "rawMarkdown": "Do I need to submit a notebook along with the prediction.csv? The 'Submit Prediction' button does not work. Is there an alternative way of submitting predictions? Patiently waiting for a response." }, { "id": 3310649, "postDate": "2025-11-03T11:12:53.367Z", "content": "

\"\"\nI’d like to ask about this image — what’s going on with it? It still has a watermark.\nAlso, do we need to take the climate into consideration?

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F23193939%2Fca95e1555a93fa3d256c37770cb91859%2FID670276799.jpg?generation=1762165289757598&alt=media)\nI’d like to ask about this image — what’s going on with it? It still has a watermark.\nAlso, do we need to take the climate into consideration?", "replies": [ { "id": 3310774, "postDate": "2025-11-03T16:19:17.410Z", "content": "

I think this is the only image with date stamp. Does test contain any - if you code to remove the date stamp does it help? Good questions to ask and attempt to answer. The line shadow is more common - what to do with it? Think this is the only train image with shadow of the person taking the picture. How to treat it?

", "rawMarkdown": "I think this is the only image with date stamp. Does test contain any - if you code to remove the date stamp does it help? Good questions to ask and attempt to answer. The line shadow is more common - what to do with it? Think this is the only train image with shadow of the person taking the picture. How to treat it?", "replies": [ { "id": 3311041, "postDate": "2025-11-04T05:00:34.620Z", "content": "

Aside from this image, I’ve noticed that at least two other images also contain the photographer’s shadow. I’m wondering — was this left unintentionally, or did the organizers include them intentionally for us to process? It would be great if the organizers could clarify this.@qiyuliaocsiro

", "rawMarkdown": "Aside from this image, I’ve noticed that at least two other images also contain the photographer’s shadow. I’m wondering — was this left unintentionally, or did the organizers include them intentionally for us to process? It would be great if the organizers could clarify this.[@qiyuliaocsiro](url)" }, { "id": 3311042, "postDate": "2025-11-04T05:06:48.167Z", "content": "

My first attempt to reduce the shadows was a bust - images look better but the cv ended up almost the same for a quick B0 5 fold model.

", "rawMarkdown": "My first attempt to reduce the shadows was a bust - images look better but the cv ended up almost the same for a quick B0 5 fold model." }, { "id": 3311093, "postDate": "2025-11-04T08:39:02.357Z", "content": "

Thank you for your careful observation and information. My answer would be neither. It only reflects the realistic conditions likely encountered in real-world applications.

\n

Unless we use a controlled experimental environment—which is not our intention—we cannot capture ground images with perfectly uniform lighting. Natural grassland settings may present various unavoidable conditions, including shadows.

\n

Furthermore, end-users are typically neither photographers nor computer vision experts. For wide deployment, it is impractical to expect all users to undergo complex training to understand and avoid all scenarios that are challenging for visual models.

", "rawMarkdown": "Thank you for your careful observation and information. My answer would be neither. It only reflects the realistic conditions likely encountered in real-world applications.\n\nUnless we use a controlled experimental environment—which is not our intention—we cannot capture ground images with perfectly uniform lighting. Natural grassland settings may present various unavoidable conditions, including shadows.\n\nFurthermore, end-users are typically neither photographers nor computer vision experts. For wide deployment, it is impractical to expect all users to undergo complex training to understand and avoid all scenarios that are challenging for visual models." }, { "id": 3311295, "postDate": "2025-11-04T16:38:21.613Z", "content": "

Just as a note of disagreement with your experimental design decision. The shadows are persons, tripods for the camera or edges of the test frame. They don't represent issues with lighting uniformity, that might occur with photos at different times of the day or weather conditions.

\n

I disagree with what seems to be a common thought with many AI experiments I see on Kaggle or elsewhere. The thought seems to be that we can collect sloppy data and let the data scientist fix it; with an assumption that it leads to better generalization and accounts for end-users being poor data collectors. I would always suggest a design that collects accurate data and than let the data scientist do transforms that lead to generalization and to account for user skills. The balance for this would move based on the planned sample size. IMO with 357 photographs on 19 locations, 3 years of data collection, several camera types and 4 seasons of the year, your design for photos should have been similar to that apparently used for NDVI data - seems like 100 measurements taken per plot was the process. That is, more efforts to keep camera and person shadows out of the frame.

\n

If I am wrong in this belief than a high scoring (0.90 or better) model will emerge in three months. If I am on the right track than the best model will be in the 0.75 range. My best model so far has given me a training CV of 0.96, but a validation CV of only 0.59.

\n

As a point of reference I have a very good friend who has a 25,000 acre (10,100 hectares) cattle ranch on the eastern plains of Colorado where rain fall is always inadequate and highly variable, resulting in a huge need to manage grazing, herd size and grass mix in each paddock very carefully. He would be an idea user of an app with this data. He's a graduate of a 4 year university degree in agricultural science and capable of taking 357 photos with much higher quality than our data. IMO ranchers not smart enough to take good photos are not still in the business.

", "rawMarkdown": "Just as a note of disagreement with your experimental design decision. The shadows are persons, tripods for the camera or edges of the test frame. They don't represent issues with lighting uniformity, that might occur with photos at different times of the day or weather conditions.\n\nI disagree with what seems to be a common thought with many AI experiments I see on Kaggle or elsewhere. The thought seems to be that we can collect sloppy data and let the data scientist fix it; with an assumption that it leads to better generalization and accounts for end-users being poor data collectors. I would always suggest a design that collects accurate data and than let the data scientist do transforms that lead to generalization and to account for user skills. The balance for this would move based on the planned sample size. IMO with 357 photographs on 19 locations, 3 years of data collection, several camera types and 4 seasons of the year, your design for photos should have been similar to that apparently used for NDVI data - seems like 100 measurements taken per plot was the process. That is, more efforts to keep camera and person shadows out of the frame.\n\nIf I am wrong in this belief than a high scoring (0.90 or better) model will emerge in three months. If I am on the right track than the best model will be in the 0.75 range. My best model so far has given me a training CV of 0.96, but a validation CV of only 0.59. \n\nAs a point of reference I have a very good friend who has a 25,000 acre (10,100 hectares) cattle ranch on the eastern plains of Colorado where rain fall is always inadequate and highly variable, resulting in a huge need to manage grazing, herd size and grass mix in each paddock very carefully. He would be an idea user of an app with this data. He's a graduate of a 4 year university degree in agricultural science and capable of taking 357 photos with much higher quality than our data. IMO ranchers not smart enough to take good photos are not still in the business.\n\n\n\n\n\n", "votes": 2 }, { "id": 3335232, "postDate": "2025-11-17T20:57:43.973Z", "content": "

This isnt just a challenge to predict biomass, it's to create a pipeline that is also robust enough to handle non vegetation on occasion without destroying prediciton entirely. The non vegetation elements are obviously placed there by design to make things interesting and represent real world examples.

", "rawMarkdown": "This isnt just a challenge to predict biomass, it's to create a pipeline that is also robust enough to handle non vegetation on occasion without destroying prediciton entirely. The non vegetation elements are obviously placed there by design to make things interesting and represent real world examples. ", "votes": 1 }, { "id": 3335257, "postDate": "2025-11-17T21:36:41.983Z", "content": "

Adam - are you telling me that you were involved in the study a decade ago? If you took the photos - we need to team up! I assume you have more :) Did notice that your avatar contains both sheep and grass!

\n

Because to the best of my knowledge 2014 kind of marks the first decent vision tools. Reading the paper for this project it shows zero intent to use vision models. Pretty sure that in 2015 anyone doing vision modeling did not obviously place crap in images to make things interesting (improve generalization). That's a much more recent concept, which you may notice from my comments I think is BS :)

", "rawMarkdown": "[Adam](https://www.kaggle.com/alloyd256) - are you telling me that you were involved in the study a decade ago? If you took the photos - we need to team up! I assume you have more :) Did notice that your avatar contains both sheep and grass!\n\nBecause to the best of my knowledge 2014 kind of marks the first decent vision tools. Reading the paper for this project it shows zero intent to use vision models. Pretty sure that in 2015 anyone doing vision modeling did not obviously place crap in images to make things interesting (improve generalization). That's a much more recent concept, which you may notice from my comments I think is BS :)", "votes": 1 } ] } ] }, { "id": 3309797, "postDate": "2025-11-01T12:34:06.647Z", "content": "

Is the test data missing something? When I downloaded it, it contained only 5-6 records and 1 image without any numerical features.

", "rawMarkdown": "Is the test data missing something? When I downloaded it, it contained only 5-6 records and 1 image without any numerical features. ", "replies": [ { "id": 3309802, "postDate": "2025-11-01T12:57:54.950Z", "content": "

that is a sample test set , actually is hidden and inferred during submission

", "rawMarkdown": "that is a sample test set , actually is hidden and inferred during submission", "replies": [ { "id": 3311815, "postDate": "2025-11-05T20:38:41.887Z", "content": "

Why does that test set only appear after we train our model with the data they give us?

", "rawMarkdown": "Why does that test set only appear after we train our model with the data they give us?" } ] } ] }, { "id": 3309057, "postDate": "2025-10-30T20:49:26.853Z", "content": "

Hello , I am wondering how can I have the train dataset. Any advice on this?

", "rawMarkdown": "Hello , I am wondering how can I have the train dataset. Any advice on this?", "replies": [ { "id": 3309061, "postDate": "2025-10-30T21:32:16.857Z", "content": "

Just go to data section in this competition and scroll down there you can find the datasets. You can click on download all if you want to use then from your local machine. But it is advisable to use it directly from kaggle, to do that\nCreate your notebook and there on the right side you can see add inputs option select that and there filter out competition datasets and search for this competition and add it by clicking on the+ symbol

\n

Hope it solves your doubt

", "rawMarkdown": "Just go to data section in this competition and scroll down there you can find the datasets. You can click on download all if you want to use then from your local machine. But it is advisable to use it directly from kaggle, to do that\nCreate your notebook and there on the right side you can see add inputs option select that and there filter out competition datasets and search for this competition and add it by clicking on the+ symbol\n\nHope it solves your doubt", "votes": 1 } ] }, { "id": 3308983, "postDate": "2025-10-30T16:39:43.110Z", "content": "

Regarding the Sampling_Date variable, is it advisable to use cyclic coding (sin/cos) to represent the month and day of the year in order to capture seasonal patterns in biomass, or is there another date feature extraction technique that is considered more appropriate in this context?

", "rawMarkdown": "Regarding the Sampling_Date variable, is it advisable to use cyclic coding (sin/cos) to represent the month and day of the year in order to capture seasonal patterns in biomass, or is there another date feature extraction technique that is considered more appropriate in this context?", "replies": [ { "id": 3309083, "postDate": "2025-10-30T23:07:32.570Z", "content": "

Hi there, you can use any pattern that you think is suitable, e.g., season, month in a circle, date cyclic coding, etc. No specific suggestion from us.

", "rawMarkdown": "Hi there, you can use any pattern that you think is suitable, e.g., season, month in a circle, date cyclic coding, etc. No specific suggestion from us." } ] }, { "id": 3308860, "postDate": "2025-10-30T12:15:47.383Z", "content": "

Could you please provide the size of the actual test dataset to calculate the inference time?

", "rawMarkdown": "Could you please provide the size of the actual test dataset to calculate the inference time?", "replies": [ { "id": 3308955, "postDate": "2025-10-30T15:46:47.877Z", "content": "

In the Data tab it indicates 800 to be the size of the test set.

", "rawMarkdown": "In the Data tab it indicates 800 to be the size of the test set.", "replies": [ { "id": 3313412, "postDate": "2025-11-09T08:18:36.847Z", "content": "

Es decir entrenamos con 300 y evaluamos con 800

", "rawMarkdown": "Es decir entrenamos con 300 y evaluamos con 800" } ] } ] }, { "id": 3308856, "postDate": "2025-10-30T12:12:48.637Z", "content": "

Hello , how to get the dataset please

", "rawMarkdown": "Hello , how to get the dataset please " }, { "id": 3308842, "postDate": "2025-10-30T11:35:04.537Z", "content": "

hello, can we use metadata like NDVI or other fileds during training??

", "rawMarkdown": "hello, can we use metadata like NDVI or other fileds during training??", "replies": [ { "id": 3309081, "postDate": "2025-10-30T23:04:02.263Z", "content": "

Yes you can, but it will not be provided in the evaluation phase

", "rawMarkdown": "Yes you can, but it will not be provided in the evaluation phase", "replies": [ { "id": 3323497, "postDate": "2025-11-14T08:44:44.853Z", "content": "

How do we construct the corresponding NDVI features for our test set? Should we treat NDVI as a separate prediction target during the training phase? Or should we construct corresponding features for the test set through other methods?

", "rawMarkdown": "How do we construct the corresponding NDVI features for our test set? Should we treat NDVI as a separate prediction target during the training phase? Or should we construct corresponding features for the test set through other methods?" }, { "id": 3323732, "postDate": "2025-11-14T12:32:17.163Z", "content": "

I think hte following page could be a good reference:\nhttps://www.kaggle.com/competitions/csiro-biomass/discussion/619477

", "rawMarkdown": "I think hte following page could be a good reference:\nhttps://www.kaggle.com/competitions/csiro-biomass/discussion/619477" } ] } ] }, { "id": 3308647, "postDate": "2025-10-29T21:24:08.100Z", "content": "

Sorry. I am not able to submit the submission.csv. Can someone help me with it?

", "rawMarkdown": "Sorry. I am not able to submit the submission.csv. Can someone help me with it?", "replies": [ { "id": 3308931, "postDate": "2025-10-30T15:04:42.050Z", "content": "

Hi, you'll need to run your notebook on Kaggle, then save a version of it.\nOpening this saved version should allow you to submit via the Output tab.

", "rawMarkdown": "Hi, you'll need to run your notebook on Kaggle, then save a version of it.\nOpening this saved version should allow you to submit via the Output tab." } ] }, { "id": 3308488, "postDate": "2025-10-29T14:47:37.400Z", "content": "

Hello,

\n

i have a question regarding: Dry green vegetation (excluding clover)\nShould we filter images that have \"Species\" == \"clover?\nIf yes what value should the predicted Dry green vegetation have?

\n

thank you in advance

", "rawMarkdown": "Hello, \n\ni have a question regarding: Dry green vegetation (excluding clover)\nShould we filter images that have \"Species\" == \"clover?\nIf yes what value should the predicted Dry green vegetation have?\n\nthank you in advance", "replies": [ { "id": 3308650, "postDate": "2025-10-29T22:14:17.547Z", "content": "

Hi there,

\n

\"Dry green vegetation\" here refers to the weight of green vegetation (excluding clover) after artificial drying. The \"Species\" field indicates the dominant plant species (one or two) present in the image, but there is no strict correlation between the two.

\n

For example:

\n

If \"Species\" does not include clover, it is still possible for \"Dry Clover\" to have a small value, which is acceptable.

\n

Conversely, if \"Species\" is labeled as \"clover,\" this does not mean that no other grasses are present in the image (they may simply not be the dominant species). In such cases, \"Dry green vegetation\" may still have a non-zero value and should be predicted accordingly.

\n

I hope this clarifies the question.

", "rawMarkdown": "Hi there,\n\n\"Dry green vegetation\" here refers to the weight of green vegetation (excluding clover) after artificial drying. The \"Species\" field indicates the dominant plant species (one or two) present in the image, but there is no strict correlation between the two.\n\nFor example:\n\nIf \"Species\" does not include clover, it is still possible for \"Dry Clover\" to have a small value, which is acceptable.\n\nConversely, if \"Species\" is labeled as \"clover,\" this does not mean that no other grasses are present in the image (they may simply not be the dominant species). In such cases, \"Dry green vegetation\" may still have a non-zero value and should be predicted accordingly.\n\nI hope this clarifies the question.", "votes": 1, "replies": [ { "id": 3310417, "postDate": "2025-11-02T21:11:52.230Z", "content": "

can I ask, why does \"Dry green vegetation\" exclude clover? Why does\"Dry clover biomass\" have it's own value?

", "rawMarkdown": "can I ask, why does \"Dry green vegetation\" exclude clover? Why does\"Dry clover biomass\" have it's own value?" }, { "id": 3310422, "postDate": "2025-11-02T21:49:58.623Z", "content": "

Tekkuitt22

\n

The hosts may provide a good explanation but I believe the clover mix is important as clover rebuilds nitrogen in the ground reducing the need to fertilize. If your grass has no clover than more fertilizer would be needed to increase the biomass.

\n

One article.

", "rawMarkdown": "[Tekkuitt22](https://www.kaggle.com/telliott22)\n\nThe hosts may provide a good explanation but I believe the clover mix is important as clover rebuilds nitrogen in the ground reducing the need to fertilize. If your grass has no clover than more fertilizer would be needed to increase the biomass.\n\n[One article.](https://www.sciencedaily.com/releases/2012/09/120927091528.htm)" }, { "id": 3310424, "postDate": "2025-11-02T21:54:03.503Z", "content": "

Question 1: This concept is defined as such.\nQuestion 2: For livestock, clover provides better taste and nutrition than ordinary vegetation. It's also important to know the structure of the biomass.

", "rawMarkdown": "Question 1: This concept is defined as such.\nQuestion 2: For livestock, clover provides better taste and nutrition than ordinary vegetation. It's also important to know the structure of the biomass." }, { "id": 3310440, "postDate": "2025-11-02T23:12:24.913Z", "content": "

I would add to @pcjimmmy and @qiyuliaocsiro that, in addition to clover being a legume (adds nitrogen and generally has a higher protiein % / digestibilty), knowing the species composition of a pasture (genetic potential) underpins accurate prediction of future feed production / quality.

\n

In biological terms, phenotype (observed pasture) is the result of the interaction between genotype and environment. Therefore, it's hard to predict future states without having a good understanding of both the genetic potential of a pasture (species composition and current biomass) and a reliable forecast of its environment (weather, grazing pressure, etc). :)

", "rawMarkdown": "I would add to @pcjimmmy and @qiyuliaocsiro that, in addition to clover being a legume (adds nitrogen and generally has a higher protiein % / digestibilty), knowing the species composition of a pasture (genetic potential) underpins accurate prediction of future feed production / quality. \n\nIn biological terms, phenotype (observed pasture) is the result of the interaction between genotype and environment. Therefore, it's hard to predict future states without having a good understanding of both the genetic potential of a pasture (species composition and current biomass) and a reliable forecast of its environment (weather, grazing pressure, etc). :)" }, { "id": 3311276, "postDate": "2025-11-04T15:52:52.557Z", "content": "

As metadata is not available for the evaluation phase, how to tackle \"\"\" reliable forecast of its environment (weather, grazing pressure, etc)\"\"\".

", "rawMarkdown": "As metadata is not available for the evaluation phase, how to tackle \"\"\" reliable forecast of its environment (weather, grazing pressure, etc)\"\"\". " }, { "id": 3311335, "postDate": "2025-11-04T18:44:58.833Z", "content": "

Training a model now to predict the metadata from the images. Looks like it might have decent cv. But the paper does not seem to suggest that tabular data created a model any better than our current LB - but only a quick read of its results. So I guess I will be adding that approach to my todo list.

", "rawMarkdown": "Training a model now to predict the metadata from the images. Looks like it might have decent cv. But the paper does not seem to suggest that tabular data created a model any better than our current LB - but only a quick read of its results. So I guess I will be adding that approach to my todo list. " }, { "id": 3311445, "postDate": "2025-11-05T03:19:03.590Z", "content": "

This challenge is focused only on how well a model is able to estimate the current state. My reference to 'predict future states' relates to how machine vision could be used to provide 'ground-truthed' data for predicting pasture growth. Apologies for any elevated cortisol :)

", "rawMarkdown": "This challenge is focused only on how well a model is able to estimate the current state. My reference to 'predict future states' relates to how machine vision could be used to provide 'ground-truthed' data for predicting pasture growth. Apologies for any elevated cortisol :)" } ] } ] }, { "id": 3309944, "postDate": "2025-11-01T18:51:47.870Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 3381378, "author_name": "Simon Alerdic", "author_url": "", "post_date": "2025-12-24T11:50:20.677000", "content": "

Thank you for organizing this interesting competition. I would like to share some feedback and suggestions regarding the prize pool distribution. Currently, the prizes are listed as:

\n
1st Place – $50,000  \n2nd Place – $20,000  \n3rd Place – $5,000  \nTotal: $75,000\n
\n

Based on the current leaderboard and related discussions, it appears that significant rank fluctuations are likely. In such a scenario, the current prize distribution may feel unfair to participants finishing in 4th or 5th place, especially if their scores are very close to those of the top three. It is also possible that the performance difference between the 1st- and 2nd-place models may not be substantial enough to justify such a large gap in rewards.

\n

I understand that this structure follows common Kaggle practices; however, given the competitive nature and tight score margins in this competition, it may warrant reconsideration. I would like to propose the following alternatives:

\n

Suggestion 1: Distribute the total prize pool among the top 5 or top 10 leaderboard positions. This approach would be both fairer and more aligned with the competitive dynamics of the leaderboard, while remaining Kaggle-friendly. For example:

\n
1st Place – $20,000  \n2nd Place – $15,000  \n3rd Place – $10,000  \n4th Place - ..\n
\n

Suggestion 2: After the competition concludes, use the final leaderboard scores of the top 5 or top 10 teams to guide a more nuanced prize distribution. For example, if the top scores are very close (e.g., 70%, 65%, and 64%), the rewards for adjacent ranks, such as 2nd and 3rd place should not differ significantly.

\n

I believe either approach would better reflect the actual performance differences and encourage broader participation.

", "votes": 3, "replies": [] }, { "id": 3370030, "author_name": "Shivam Shinde", "author_url": "", "post_date": "2025-12-10T14:29:57.190000", "content": "

Hi @qiyuliaocsiro, Can we use DINOv3 pretrained models? Its license is not OSI-approved and not a GPL3 license, but Section 2.5.a mentions exceptions for pretrained models with incompatible licenses. Please clarify if this applies here.

\n

License reference: https://github.com/facebookresearch/dinov3/blob/main/LICENSE.md

", "votes": 3, "replies": [ { "id": 3370575, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-12-10T23:58:42.047000", "content": "

Hi Shivam, please refer to this topic: https://www.kaggle.com/competitions/csiro-biomass/discussion/617603#3322386

", "votes": 0, "replies": [ { "id": 3374536, "author_name": "Naitik Kariwal", "author_url": "", "post_date": "2025-12-12T18:18:13.657000", "content": "

Hey @qiyuliaocsiro,

\n
    \n
  1. Is the weighted r2 computed as per Kaggle description (uses global ground truth mean and weighting in denominator) or the attached literature (computes R2 for each biomass component separately). Both result in different results.
    2. \n
  2. Are you evaluating our result on log-transformed values?
    3. \n
", "votes": 0, "replies": [] }, { "id": 3377199, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-12-15T23:34:58.833000", "content": "

Hi Naitik, Everything is subject to the Kaggle description page.

", "votes": 0, "replies": [] } ] } ] }, { "id": 3387310, "author_name": "JROOT", "author_url": "", "post_date": "2026-01-06T17:54:16.560000", "content": "

Hi @qiyuliaocsiro,

\n

I have a question regarding the use of test data during the inference phase.

\n

Since this is a Code Competition, are we permitted to use unsupervised techniques on the test set during the notebook runtime? Specifically, I am interested in:

\n

Transductive / Continual Learning: Updating model weights or batch normalization statistics dynamically using the test set features (without labels) during the inference run.

\n

Post-processing: adjusting final predictions based on the statistical distribution of the test data batch.

\n

I want to ensure my approach remains within the competition rules regarding data leakage.

", "votes": 1, "replies": [ { "id": 3387839, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-07T17:32:09.027000", "content": "

Not an official answer but working on that exact thing today :) When I have tried it in the past Compute Time has always been an issue, in fact first attempt last night ran over the time limit and died this morning.

", "votes": 0, "replies": [] } ] }, { "id": 3364771, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-12-06T17:44:54.717000", "content": "

Qiyu

\n

Its not clear to me how the evaluation handled weeds in the study.

\n

For example, Crumbweed is bad - too much can make animals sick. One would assume that you do not want it listed as Biomass - but…

\n

Were known weeds not included in the grass or clover or dead values?

", "votes": 1, "replies": [ { "id": 3370573, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-12-10T23:56:32.213000", "content": "

Hi Jimmy, that's another critical research topic. I think we'll have some following work for that.

\n

For now, we only consider the distinction between green grass and dead clover. Finer classifications, such as separating out specific weeds like crumbweed, are currently not feasible.

", "votes": 0, "replies": [] } ] }, { "id": 3359489, "author_name": "khushirani22", "author_url": "", "post_date": "2025-12-01T18:16:32.843000", "content": "

I think the current dataset is excellent for one specific task:\ngiven an image + NDVI + height + date + species, predict the biomass at that moment.\nThat’s a well-defined ML problem, and many models will perform strongly on it.

\n

However, real grazing and pasture management are dynamic and path-dependent systems. Two paddocks with the same present biomass can behave very differently depending on their recent grazing pressure. Rainfall history, stress cycles, soil type, paddock identity, and past management all influence recovery and biomass regrowth has real hysteresis. Overgrazed paddocks do not recover like rested ones.

\n

None of this structure is represented in the dataset. There are no paddock IDs, no longitudinal sequences, no stocking density or grazing events, no soil properties, and no management history ,only static snapshots treated as independent samples. As a result, NDVI + height + seasonal timing already capture most of the explainable signal, and image texture provides some marginal improvement, but true pasture recovery dynamics are nearly impossible to learn from this format.

\n

For a future version, it would be incredibly valuable to include repeated measurements from the same paddock over time, explicit grazing/rest intervals, rainfall summaries, soil type, and stocking history. That would allow models to answer the questions farmers genuinely care about, not just “How much biomass is here today?” but “How will this paddock respond if I graze it now versus five days later?”

\n

Thanks to the organizers !!

", "votes": 1, "replies": [ { "id": 3360212, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-12-02T01:18:26.543000", "content": "

I believe building a comprehensive multimodal model to directly address the practical questions farmers face and support reliable decision-making is our ultimate goal.

\n

As you can see, we are still some distance from achieving that in this research aera: real-world data is fragmented, difficult to align and integrate, and certain fundamental data dimensions are challenging to collect. Computer vision–based biomass estimation is an important step toward that vision—it enables large-scale, low-cost, and pottentially accurate biomass measurement. Only with such capability can we begin to meaningfully connect biomass with other factors and make robust management decisions.

\n

Thank you for your thoughtful comment. Your suggestions are highly valuable, and we will certainly consider them in our future work.

", "votes": 1, "replies": [] }, { "id": 3369917, "author_name": "Ali Shakir", "author_url": "", "post_date": "2025-12-10T12:57:08.993000", "content": "

I thought it was to give an image only -> then predict NDVI + height + date + species + target

", "votes": 0, "replies": [] } ] }, { "id": 3334673, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-11-17T15:53:20.933000", "content": "

In an attempt to understand the data I have tried to model clover and grass growth over the year. My models don't make sense! Do you know if the pastures used in the data collection were used for grazing during 2015?

\n

That would seem to be the most likely cause for my issues.

", "votes": 1, "replies": [ { "id": 3334801, "author_name": "Muhammad Awais", "author_url": "", "post_date": "2025-11-17T16:34:54.040000", "content": "

I also just confused, why they include year column.

", "votes": 0, "replies": [] }, { "id": 3335547, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-11-18T04:51:43.937000", "content": "

I believe the grazing status of each sampling site can no longer be traced back and is, in fact, difficult to quantify.

", "votes": 0, "replies": [] } ] }, { "id": 3313558, "author_name": "Shivam Shinde", "author_url": "", "post_date": "2025-11-09T13:14:33.047000", "content": "

Hi @qiyuliaocsiro ,\nCan we pretrain a model locally on additional datasets that are publicly available online?\nWould using these locally pretrained model weights for the official submission be allowed, or would it count as using external data against the competition rules?

", "votes": 1, "replies": [ { "id": 3313856, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-11-09T23:13:03.470000", "content": "

Yes, you are allowed to use any public datasets that are also available to other competitors, as stated in Rule 2.6.

", "votes": 0, "replies": [] } ] }, { "id": 3308587, "author_name": "AYUSHBHALLA", "author_url": "", "post_date": "2025-10-29T17:50:08.673000", "content": "

NDVI data's format? Per-image value or spatial map?

", "votes": 1, "replies": [ { "id": 3308661, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-10-29T23:25:12.983000", "content": "

Per-image value

", "votes": 0, "replies": [] } ] }, { "id": 3308325, "author_name": "Manoj", "author_url": "", "post_date": "2025-10-29T05:27:34.420000", "content": "

i am guessing this will have massive participation

", "votes": 1, "replies": [] }, { "id": 3386151, "author_name": "Muhammad", "author_url": "", "post_date": "2026-01-04T17:32:39.133000", "content": "

Hello, could anyone please help me check why the code runs successfully, but the submission fails.

\n

\"\"

", "votes": -1, "replies": [ { "id": 3386292, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-05T01:43:53.037000", "content": "

Put a link to your notebook after you make it public - makes it easier for folks to help. Without the notebook - suggest you read all the posts - use 'submission' in the search and you will see a bunch.

", "votes": 0, "replies": [ { "id": 3387332, "author_name": "Muhammad", "author_url": "", "post_date": "2026-01-06T18:36:18.603000", "content": "

here is the link: https://www.kaggle.com/code/ammaruk/csiro-biomass-error-fix?scriptVersionId=288706820

\n

Thanks for helping

", "votes": 0, "replies": [] }, { "id": 3387350, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-06T19:17:30.243000", "content": "

The link is broken. I looked and don't see any public shared notebook by you.

", "votes": 0, "replies": [] }, { "id": 3390513, "author_name": "Muhammad", "author_url": "", "post_date": "2026-01-13T10:10:49.320000", "content": "

I have made the book public, sorry I didn't know that.

\n

https://www.kaggle.com/code/ammaruk/csiro-biomass-error-fix/edit/run/288706820

", "votes": 0, "replies": [] }, { "id": 3390546, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-13T11:06:42.993000", "content": "

Ok - can see your notebook now. But I cannot fork and run the notebook as your data sets are private. You will need to also make those public.

\n

It does seem that your doing the embedding on local machine and reading those from a file. You cannot embed the test images on local machine. As I read the code it does seem like your only going to do a single image.

\n

Have your tried predicting on the train data

\n
train_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/test.csv')\n
\n

change to

\n
train_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\n
\n

Common mistake is that the code will only handle the single test image - predicting on the train data exposes those errors.

", "votes": 0, "replies": [] }, { "id": 3390563, "author_name": "Muhammad", "author_url": "", "post_date": "2026-01-13T11:48:31.133000", "content": "

Actually I embedded the train images on Kaggle but then I got submission error and made a new notebook in which I removed the train_img_embedding (coding bit) because I downloaded the embeddings and uploaded them from local computer. Anyways, I have made the datasets public.

\n

Btw, if I am not putting the train_img_embedding code bit, can it effect my submission?

\n

Thanks for pointing it out: test_data = pd.read_csv('/kaggle/input/csiro-biomass/test.csv')

\n

I will change it.

", "votes": 0, "replies": [] }, { "id": 3390625, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-13T14:48:28.173000", "content": "

Guess I do not understand your test embeddings - assume you did that on local machine. Are they the result of the train images or did they come from the single test image we see?

\n

Forking and running your code now attempting to predict on train rather than test.

", "votes": 0, "replies": [] }, { "id": 3390628, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-13T14:50:21.443000", "content": "

So - if you replace test.csv with train.csv and run the code in the editor you quickly get to your error. Your only generating a single image worth in the dataframe. This generates a shape issue a cell or two later.

\n
train_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\ntest_data = pd.read_csv('/kaggle/input/csiro-biomass/train.csv')\n
", "votes": 0, "replies": [] } ] }, { "id": 3396977, "author_name": "", "author_url": "", "post_date": "2026-01-26T08:27:14.600000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3381440, "author_name": "Khushi Yadav", "author_url": "", "post_date": "2025-12-24T15:47:45.540000", "content": "

Image2Biomass Prediction

", "votes": -10, "replies": [] }, { "id": 3399136, "author_name": "Aic", "author_url": "", "post_date": "2026-01-30T09:05:01.703000", "content": "

Hello @qiyuliaocsiro

\n

Why have my ranking and score been cancelled? I did not violate any rules during this competition, and I open-sourced my code in the early stages. Could the organizers please explain why I can no longer find my ranking? Thank you very much.

", "votes": 0, "replies": [] }, { "id": 3399015, "author_name": "Yiyang_Hu", "author_url": "", "post_date": "2026-01-30T04:02:35.833000", "content": "

Why have my competition results disappeared, and why does it show that I did not participate in this competition?@addisonhoward @qiyuliaocsiro

", "votes": 0, "replies": [] }, { "id": 3397128, "author_name": "Yiyang_Hu", "author_url": "", "post_date": "2026-01-26T14:44:23.377000", "content": "

“”You've exceeded the number of hours allowed to use GPU. Your quota will reset to 30 hours in 4 days. You can continue working in this session until you've reached your 9 hour limit. You will not be able to commit unless you turn off GPU.“” how could i submit with gpu？

", "votes": 0, "replies": [] }, { "id": 3392937, "author_name": "Leiner Barba", "author_url": "", "post_date": "2026-01-17T21:16:06.110000", "content": "

Hi, I have some questions. The train.csv file contains the values of NVDI, height and others, associated to each image of the dataset. However, the test.csv does not have those values. I mean, during scoring, will these values be provided? Should we avoid to build the model using those values?

", "votes": 0, "replies": [ { "id": 3392967, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-17T23:54:20.777000", "content": "

They will not be provided. If you want to use them you will need to create them, use means/medians or not use them.

", "votes": 0, "replies": [] } ] }, { "id": 3392419, "author_name": "Mohamed Drabo", "author_url": "", "post_date": "2026-01-16T23:38:48.190000", "content": "

What's wrong with submission file get keep get this error : Notebook Threw Exception · 21m ago \nthis the link to y code : https://www.kaggle.com/code/mohameddrabo/csiro-image2biomass-prediction

", "votes": 0, "replies": [ { "id": 3392423, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-16T23:49:43", "content": "

Lots of discussion posts on this error. I even have one. This will not work

\n
sample  = pd.read_csv('/kaggle/input/csiro-biomass/sample_submission.csv')\n
\n

Because kaggle screwed up (IMO) this file is not available in the private test folder when scoring submission. Your code will likely work if you use test.csv instead.

", "votes": 0, "replies": [] } ] }, { "id": 3391634, "author_name": "Nebou Rocket", "author_url": "", "post_date": "2026-01-15T10:47:27.207000", "content": "

How should I use other features?

", "votes": 0, "replies": [] }, { "id": 3390384, "author_name": "yiyang hu", "author_url": "", "post_date": "2026-01-13T04:02:57.150000", "content": "

i met this problem“Couldn't verify your identity Persona couldn't verify your identity. If you feel this is in error, contact support.” all my submission disappeared，and i got unranked，but i got 243th. can you help me solve this issue?

", "votes": 0, "replies": [] }, { "id": 3390382, "author_name": "yiyang hu", "author_url": "", "post_date": "2026-01-13T04:00:56.520000", "content": "

Hello, I would like to inquire about an issue with my account. My competition rankings and submission records have disappeared. Furthermore, the rankings from my past competitions have all changed to 'unrank'. Could you please look into this and resolve the issue?

", "votes": 0, "replies": [ { "id": 3390383, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2026-01-13T04:02:24.927000", "content": "

Probably not the best place to ask for this help.

", "votes": 0, "replies": [] } ] }, { "id": 3381180, "author_name": "Abrar Sadikeen", "author_url": "", "post_date": "2025-12-24T01:06:20.677000", "content": "

Hi! I'm not sure I understand the evaluation criteria. Is it R2 of the entire train+test set using only images, or R2 of the test set using only images (contains 1 image)?

", "votes": 0, "replies": [ { "id": 3381186, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-12-24T01:33:08.577000", "content": "

The test set that will be evaluated is unseen to us - right now only a single dummy image is in test - when you submit for a score it will be replaced by around 800 images. The train images are not evaluated for the LB score.

", "votes": 0, "replies": [] } ] }, { "id": 3378520, "author_name": "Yiyang_Hu", "author_url": "", "post_date": "2025-12-18T12:22:06.773000", "content": "

Hello, could you please help me check why the code runs successfully, but the submission fails and no score is obtained?<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5396927%2Fb8326621c34235dc04963051d5d8e63b%2F2.png?generation=1766060523804972&alt=media\" alt=\"! (opens in a new tab)\">\" />

", "votes": 0, "replies": [ { "id": 3378521, "author_name": "Yiyang_Hu", "author_url": "", "post_date": "2025-12-18T12:22:51.327000", "content": "

\"\"

", "votes": 0, "replies": [ { "id": 3378964, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-12-18T22:54:28.030000", "content": "

Couple of common reasons in this competition - 1) You cannot read sample_submission.csv as for some reason it does not exist in the private test folder 2) you have a shape issue that does not show up with single image in the public test.

\n

change to read the train.csv instead of test.csv so that more images are predicted. You will almost certainly see an error issue you can act on - fix that and than go back to reading test.

\n

That does not do the trick - make your notebook public and share a link here. Someone might spot the issue.

", "votes": 0, "replies": [] } ] } ] }, { "id": 3378490, "author_name": "Aic", "author_url": "", "post_date": "2025-12-18T10:54:20.697000", "content": "

@qiyuliaocsiro Since the test set is split by time, I was wondering if the time distribution between the Public and Private Leaderboards is uniform (i.e., do they both cover all four seasons)? Additionally, is the species distribution balanced between the two? I would be extremely grateful for your reply!

", "votes": 0, "replies": [] }, { "id": 3375145, "author_name": "Muhammad", "author_url": "", "post_date": "2025-12-12T21:04:25.683000", "content": "

Hi @qiyuliaocsiro, I am getting a Notebook Threw Exception error. Can you suggest me any ideas or solutions to debug it. Also, do you have any ideas what type of error it could be in Notebook Threw Exception. Thanks!🙏

", "votes": 0, "replies": [ { "id": 3377200, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-12-15T23:37:18.887000", "content": "

Sorry, I'm not able to resolve any operational issues. Please contact the Kaggle official for support

", "votes": 0, "replies": [] } ] }, { "id": 3371266, "author_name": "Ali Shakir", "author_url": "", "post_date": "2025-12-11T11:36:17.870000", "content": "

Hello,

\n

Grass Weight Prediction

\n

What is this?

\n

Predict how much grass weighs from photos of pastures.

\n

Input

\n\n

Output

\n

Predict 5 weights (in grams):

\n\n

Problem

\n

Test data only has photos, no extra info. Need to use images to predict.

\n

I do not understand. if the input are only images and that is it?

", "votes": 0, "replies": [ { "id": 3371694, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-12-11T16:00:54.337000", "content": "

You can model only the images and predict. Or you can model the extra info and the images for predictions.

", "votes": 0, "replies": [] }, { "id": 3385943, "author_name": "girish2002", "author_url": "", "post_date": "2026-01-04T09:41:45.617000", "content": "

So the task basically becomes predicting target based on the image?\nWe can not use the additional feature while training as those information are not available for test dataset.

", "votes": 0, "replies": [ { "id": 3386089, "author_name": "JROOT", "author_url": "", "post_date": "2026-01-04T15:34:45.037000", "content": "

you could train a smaller model to predict those additional features and then include that as input to the final model

", "votes": 0, "replies": [] } ] } ] }, { "id": 3348128, "author_name": "c0sm1c", "author_url": "", "post_date": "2025-11-25T16:14:15.600000", "content": "

Hello, I would like to know if the test set has the same metadata as the training set?

", "votes": 0, "replies": [ { "id": 3348215, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-11-25T17:05:52.127000", "content": "

No - there is no metadata for the test set!

", "votes": 0, "replies": [ { "id": 3349132, "author_name": "c0sm1c", "author_url": "", "post_date": "2025-11-26T13:16:10.927000", "content": "

Thank you. I understand.

", "votes": 0, "replies": [] } ] } ] }, { "id": 3344863, "author_name": "", "author_url": "", "post_date": "2025-11-23T03:14:25.280000", "content": "

today ,I don't know what's going on . users from china couldn't get on kaggle through chinese internet without vpn

", "votes": 0, "replies": [] }, { "id": 3340613, "author_name": "SoWhe", "author_url": "", "post_date": "2025-11-19T16:23:39.250000", "content": "

Dear Qiyu Laio, I am having difficultie in submitting the file. I trun off internet of the setting panel, and it seem to be another tag on the right panel; so when knitting the submit button, i am having a message that I should turn off internet; so please which one should turn?

\n

Thank you

", "votes": 0, "replies": [ { "id": 3341109, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-11-20T00:03:01.127000", "content": "

Sorry Sowhe, I can hardly help solve any submission or operation issues.\n@martynaplomecka Hi Martyna, can you help with this?

", "votes": 0, "replies": [ { "id": 3342106, "author_name": "MartynaPlomecka", "author_url": "", "post_date": "2025-11-20T16:31:58.763000", "content": "

The internet must be turned off during submission. Please make sure it is disabled when submitting

", "votes": 0, "replies": [] }, { "id": 3362432, "author_name": "", "author_url": "", "post_date": "2025-12-04T17:35:11.207000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3336617, "author_name": "Muhammad", "author_url": "", "post_date": "2025-11-18T14:54:12.050000", "content": "

Would it be better to display more digits after the decimal point in the evaluation metric score?

", "votes": 0, "replies": [] }, { "id": 3331053, "author_name": "Vasant Rajadhyax", "author_url": "", "post_date": "2025-11-16T11:14:47.727000", "content": "

Not able to submit notebook for prediction. Also there is no error to debug in logs. \nnotebook link: https://www.kaggle.com/code/vasant10/csiro-image2biomass-multi-output-regression

", "votes": 0, "replies": [ { "id": 3347437, "author_name": "Mehmet Ali KABA", "author_url": "", "post_date": "2025-11-25T06:35:47.577000", "content": "

I am having the same issue.

", "votes": 0, "replies": [] }, { "id": 3362493, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-12-04T19:03:01.620000", "content": "

Vasabt

\n

Mehmet

\n

When submissions fails but running against the small test is ok - you should ALWAYS replace test with train and see if any memory, size assumptions, timing issues exist, etc. I changed the cell for reading the test file to this:

\n

test_df = pd.read_csv(\"/kaggle/input/csiro-biomass/train.csv\")

\n

This will generate an error for you to debug. After the code will predict on train than you can change it back to test.csv and try for a submission.

\n

Good luck.

", "votes": 0, "replies": [] } ] }, { "id": 3320938, "author_name": "", "author_url": "", "post_date": "2025-11-13T00:05:16.700000", "content": "", "votes": 0, "replies": [ { "id": 3320943, "author_name": "", "author_url": "", "post_date": "2025-11-13T00:13:44.467000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 3311506, "author_name": "", "author_url": "", "post_date": "2025-11-05T06:42:54.447000", "content": "", "votes": 0, "replies": [ { "id": 3376905, "author_name": "", "author_url": "", "post_date": "2025-12-15T11:45:51.623000", "content": "", "votes": 0, "replies": [ { "id": 3377202, "author_name": "", "author_url": "", "post_date": "2025-12-15T23:38:43.953000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3311447, "author_name": "", "author_url": "", "post_date": "2025-11-05T03:27:31.223000", "content": "", "votes": 0, "replies": [ { "id": 3311854, "author_name": "", "author_url": "", "post_date": "2025-11-05T23:43:44.350000", "content": "", "votes": 3, "replies": [ { "id": 3360108, "author_name": "", "author_url": "", "post_date": "2025-12-01T21:31:21.963000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3311340, "author_name": "", "author_url": "", "post_date": "2025-11-04T19:10:18.323000", "content": "", "votes": 0, "replies": [ { "id": 3311443, "author_name": "", "author_url": "", "post_date": "2025-11-05T03:07:00.240000", "content": "", "votes": 0, "replies": [ { "id": 3362439, "author_name": "", "author_url": "", "post_date": "2025-12-04T17:38:28.190000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3311044, "author_name": "", "author_url": "", "post_date": "2025-11-04T05:16:22.230000", "content": "", "votes": 0, "replies": [ { "id": 3311046, "author_name": "", "author_url": "", "post_date": "2025-11-04T05:25:37.970000", "content": "", "votes": 0, "replies": [ { "id": 3313245, "author_name": "", "author_url": "", "post_date": "2025-11-09T02:00:40.370000", "content": "", "votes": 0, "replies": [] }, { "id": 3313335, "author_name": "", "author_url": "", "post_date": "2025-11-09T05:36:55.143000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3310894, "author_name": "", "author_url": "", "post_date": "2025-11-03T20:36:33.763000", "content": "", "votes": 0, "replies": [] }, { "id": 3310649, "author_name": "", "author_url": "", "post_date": "2025-11-03T11:12:53.367000", "content": "", "votes": 0, "replies": [ { "id": 3310774, "author_name": "", "author_url": "", "post_date": "2025-11-03T16:19:17.410000", "content": "", "votes": 0, "replies": [ { "id": 3311041, "author_name": "", "author_url": "", "post_date": "2025-11-04T05:00:34.620000", "content": "", "votes": 0, "replies": [] }, { "id": 3311042, "author_name": "", "author_url": "", "post_date": "2025-11-04T05:06:48.167000", "content": "", "votes": 0, "replies": [] }, { "id": 3311093, "author_name": "", "author_url": "", "post_date": "2025-11-04T08:39:02.357000", "content": "", "votes": 0, "replies": [] }, { "id": 3311295, "author_name": "", "author_url": "", "post_date": "2025-11-04T16:38:21.613000", "content": "", "votes": 2, "replies": [] }, { "id": 3335232, "author_name": "", "author_url": "", "post_date": "2025-11-17T20:57:43.973000", "content": "", "votes": 1, "replies": [] }, { "id": 3335257, "author_name": "", "author_url": "", "post_date": "2025-11-17T21:36:41.983000", "content": "", "votes": 1, "replies": [] } ] } ] }, { "id": 3309797, "author_name": "", "author_url": "", "post_date": "2025-11-01T12:34:06.647000", "content": "", "votes": 0, "replies": [ { "id": 3309802, "author_name": "", "author_url": "", "post_date": "2025-11-01T12:57:54.950000", "content": "", "votes": 0, "replies": [ { "id": 3311815, "author_name": "", "author_url": "", "post_date": "2025-11-05T20:38:41.887000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3309057, "author_name": "", "author_url": "", "post_date": "2025-10-30T20:49:26.853000", "content": "", "votes": 0, "replies": [ { "id": 3309061, "author_name": "", "author_url": "", "post_date": "2025-10-30T21:32:16.857000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 3308983, "author_name": "", "author_url": "", "post_date": "2025-10-30T16:39:43.110000", "content": "", "votes": 0, "replies": [ { "id": 3309083, "author_name": "", "author_url": "", "post_date": "2025-10-30T23:07:32.570000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3308860, "author_name": "", "author_url": "", "post_date": "2025-10-30T12:15:47.383000", "content": "", "votes": 0, "replies": [ { "id": 3308955, "author_name": "", "author_url": "", "post_date": "2025-10-30T15:46:47.877000", "content": "", "votes": 0, "replies": [ { "id": 3313412, "author_name": "", "author_url": "", "post_date": "2025-11-09T08:18:36.847000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3308856, "author_name": "", "author_url": "", "post_date": "2025-10-30T12:12:48.637000", "content": "", "votes": 0, "replies": [] }, { "id": 3308842, "author_name": "", "author_url": "", "post_date": "2025-10-30T11:35:04.537000", "content": "", "votes": 0, "replies": [ { "id": 3309081, "author_name": "", "author_url": "", "post_date": "2025-10-30T23:04:02.263000", "content": "", "votes": 0, "replies": [ { "id": 3323497, "author_name": "", "author_url": "", "post_date": "2025-11-14T08:44:44.853000", "content": "", "votes": 0, "replies": [] }, { "id": 3323732, "author_name": "", "author_url": "", "post_date": "2025-11-14T12:32:17.163000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3308647, "author_name": "", "author_url": "", "post_date": "2025-10-29T21:24:08.100000", "content": "", "votes": 0, "replies": [ { "id": 3308931, "author_name": "", "author_url": "", "post_date": "2025-10-30T15:04:42.050000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3308488, "author_name": "", "author_url": "", "post_date": "2025-10-29T14:47:37.400000", "content": "", "votes": 0, "replies": [ { "id": 3308650, "author_name": "", "author_url": "", "post_date": "2025-10-29T22:14:17.547000", "content": "", "votes": 1, "replies": [ { "id": 3310417, "author_name": "", "author_url": "", "post_date": "2025-11-02T21:11:52.230000", "content": "", "votes": 0, "replies": [] }, { "id": 3310422, "author_name": "", "author_url": "", "post_date": "2025-11-02T21:49:58.623000", "content": "", "votes": 0, "replies": [] }, { "id": 3310424, "author_name": "", "author_url": "", "post_date": "2025-11-02T21:54:03.503000", "content": "", "votes": 0, "replies": [] }, { "id": 3310440, "author_name": "", "author_url": "", "post_date": "2025-11-02T23:12:24.913000", "content": "", "votes": 0, "replies": [] }, { "id": 3311276, "author_name": "", "author_url": "", "post_date": "2025-11-04T15:52:52.557000", "content": "", "votes": 0, "replies": [] }, { "id": 3311335, "author_name": "", "author_url": "", "post_date": "2025-11-04T18:44:58.833000", "content": "", "votes": 0, "replies": [] }, { "id": 3311445, "author_name": "", "author_url": "", "post_date": "2025-11-05T03:19:03.590000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 3309944, "author_name": "", "author_url": "", "post_date": "2025-11-01T18:51:47.870000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3308275": "We are thrilled to host this challenge because we believe the brightest data minds from around the globe can help solve one of the most persistent and meaningful problems in modern agriculture.\n\nIf you have any questions about the data, evaluation criteria, or any other aspect of the competition, please feel free to post them in the Discussion forum. We welcome all questions and discussions, and I will do my best to answer them promptly.\n\nThank you for joining us, and we wish you the very best of luck!", "3381378": "Thank you for organizing this interesting competition. I would like to share some feedback and suggestions regarding the prize pool distribution. Currently, the prizes are listed as:\n\n```\n1st Place – $50,000 \n2nd Place – $20,000 \n3rd Place – $5,000 \nTotal: $75,000\n```\n\nBased on the current leaderboard and related discussions, it appears that significant rank fluctuations are likely. In such a scenario, the current prize distribution may feel unfair to participants finishing in `4th` or `5th` place, especially if their scores are very close to those of the top three. It is also possible that the performance difference between the 1st- and 2nd-place models may not be substantial enough to justify such a large gap in rewards.\n\nI understand that this structure follows common Kaggle practices; however, given the competitive nature and tight score margins in this competition, it may warrant reconsideration. I would like to propose the following alternatives:\n\n**Suggestion 1:** Distribute the total prize pool among the top `5` or top `10` leaderboard positions. This approach would be both fairer and more aligned with the competitive dynamics of the leaderboard, while remaining Kaggle-friendly. For example:\n\n```\n1st Place – $20,000 \n2nd Place – $15,000 \n3rd Place – $10,000 \n4th Place - ..\n```\n\n**Suggestion 2:** After the competition concludes, use the final leaderboard scores of the top `5` or top `10` teams to guide a more nuanced prize distribution. For example, if the top scores are very close (e.g., 70%, 65%, and 64%), the rewards for adjacent ranks, such as 2nd and 3rd place should not differ significantly.\n\nI believe either approach would better reflect the actual performance differences and encourage broader participation.", "3370030": "Hi @qiyuliaocsiro, Can we use DINOv3 pretrained models? Its license is not OSI-approved and not a GPL3 license, but Section 2.5.a mentions exceptions for pretrained models with incompatible licenses. Please clarify if this applies here.\n\nLicense reference: https://github.com/facebookresearch/dinov3/blob/main/LICENSE.md", "3387310": "Hi @qiyuliaocsiro,\n\nI have a question regarding the use of test data during the inference phase.\n\nSince this is a Code Competition, are we permitted to use unsupervised techniques on the test set during the notebook runtime? Specifically, I am interested in:\n\nTransductive / Continual Learning: Updating model weights or batch normalization statistics dynamically using the test set features (without labels) during the inference run.\n\nPost-processing: adjusting final predictions based on the statistical distribution of the test data batch.\n\nI want to ensure my approach remains within the competition rules regarding data leakage.", "3364771": "[Qiyu](https://www.kaggle.com/qiyuliaocsiro)\n\nIts not clear to me how the evaluation handled weeds in the study. \n\nFor example, Crumbweed is bad - too much can make animals sick. One would assume that you do not want it listed as Biomass - but...\n\nWere known weeds not included in the grass or clover or dead values?", "3359489": "I think the current dataset is excellent for one specific task:\ngiven an image + NDVI + height + date + species, predict the biomass at that moment.\nThat’s a well-defined ML problem, and many models will perform strongly on it.\n\nHowever, real grazing and pasture management are dynamic and path-dependent systems. Two paddocks with the same present biomass can behave very differently depending on their recent grazing pressure. Rainfall history, stress cycles, soil type, paddock identity, and past management all influence recovery and biomass regrowth has real hysteresis. Overgrazed paddocks do not recover like rested ones.\n\nNone of this structure is represented in the dataset. There are no paddock IDs, no longitudinal sequences, no stocking density or grazing events, no soil properties, and no management history ,only static snapshots treated as independent samples. As a result, NDVI + height + seasonal timing already capture most of the explainable signal, and image texture provides some marginal improvement, but true pasture recovery dynamics are nearly impossible to learn from this format.\n\nFor a future version, it would be incredibly valuable to include repeated measurements from the same paddock over time, explicit grazing/rest intervals, rainfall summaries, soil type, and stocking history. That would allow models to answer the questions farmers genuinely care about, not just “How much biomass is here today?” but “How will this paddock respond if I graze it now versus five days later?”\n\nThanks to the organizers !!", "3334673": "In an attempt to understand the data I have tried to model clover and grass growth over the year. My models don't make sense! Do you know if the pastures used in the data collection were used for grazing during 2015?\n\nThat would seem to be the most likely cause for my issues.", "3313558": "Hi @qiyuliaocsiro ,\nCan we pretrain a model locally on additional datasets that are **publicly available** online?\nWould using these locally **pretrained model weights** for the official submission be allowed, or would it count as using external data against the competition rules?", "3308587": "NDVI data's format? Per-image value or spatial map?", "3308325": "i am guessing this will have massive participation", "3386151": "Hello, could anyone please help me check why the code runs successfully, but the submission fails.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F27667739%2Fe74eaa772258028ca1eff97e22c11234%2FScreenshot%202026-01-04%20at%205.28.04pm.png?generation=1767547700190890&alt=media)", "3381440": "Image2Biomass Prediction", "3399136": "Hello @qiyuliaocsiro\n\nWhy have my ranking and score been cancelled? I did not violate any rules during this competition, and I open-sourced my code in the early stages. Could the organizers please explain why I can no longer find my ranking? Thank you very much.", "3399015": "Why have my competition results disappeared, and why does it show that I did not participate in this competition?@addisonhoward @qiyuliaocsiro", "3397128": "“”You've exceeded the number of hours allowed to use GPU. Your quota will reset to 30 hours in 4 days. You can continue working in this session until you've reached your 9 hour limit. You will not be able to commit unless you turn off GPU.“” how could i submit with gpu？", "3392937": "Hi, I have some questions. The train.csv file contains the values of NVDI, height and others, associated to each image of the dataset. However, the test.csv does not have those values. I mean, during scoring, will these values be provided? Should we avoid to build the model using those values?", "3392419": "What's wrong with submission file get keep get this error : Notebook Threw Exception · 21m ago \nthis the link to y code : https://www.kaggle.com/code/mohameddrabo/csiro-image2biomass-prediction", "3391634": "How should I use other features?", "3390384": "i met this problem“Couldn't verify your identity Persona couldn't verify your identity. If you feel this is in error, contact support.” all my submission disappeared，and i got unranked，but i got 243th. can you help me solve this issue?", "3390382": "Hello, I would like to inquire about an issue with my account. My competition rankings and submission records have disappeared. Furthermore, the rankings from my past competitions have all changed to 'unrank'. Could you please look into this and resolve the issue?", "3381180": "Hi! I'm not sure I understand the evaluation criteria. Is it R2 of the entire train+test set using only images, or R2 of the test set using only images (contains 1 image)? ", "3378520": "Hello, could you please help me check why the code runs successfully, but the submission fails and no score is obtained?![![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5396927%2Fb8326621c34235dc04963051d5d8e63b%2F2.png?generation=1766060523804972&alt=media)](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5396927%2F063337bde5ffd3e65b665a773e7bab7e%2F1.png?generation=1766060514514656&alt=media)", "3378490": "@qiyuliaocsiro Since the test set is split by time, I was wondering if the time distribution between the Public and Private Leaderboards is uniform (i.e., do they both cover all four seasons)? Additionally, is the species distribution balanced between the two? I would be extremely grateful for your reply!", "3375145": "Hi @qiyuliaocsiro, I am getting a Notebook Threw Exception error. Can you suggest me any ideas or solutions to debug it. Also, do you have any ideas what type of error it could be in Notebook Threw Exception. Thanks!🙏 ", "3371266": "Hello, \n# Grass Weight Prediction\n\n## What is this?\nPredict how much grass weighs from photos of pastures.\n\n## Input\n- Photos of grass fields\n- Some extra info: greenness score (NDVI), grass height, location, grass type, date\n\n## Output\nPredict 5 weights (in grams):\n- Total dry grass\n- Green grass\n- Dead grass \n- Clover\n- Fresh grass (GDM)\n\n## Problem\nTest data only has photos, no extra info. Need to use images to predict.\n\nI do not understand. if the input are only images and that is it?", "3348128": "Hello, I would like to know if the test set has the same metadata as the training set?", "3344863": "today ,I don't know what's going on . users from china couldn't get on kaggle through chinese internet without vpn", "3340613": "Dear Qiyu Laio, I am having difficultie in submitting the file. I trun off internet of the setting panel, and it seem to be another tag on the right panel; so when knitting the submit button, i am having a message that I should turn off internet; so please which one should turn?\n\nThank you", "3336617": "Would it be better to display more digits after the decimal point in the evaluation metric score?", "3331053": "Not able to submit notebook for prediction. Also there is no error to debug in logs. \nnotebook link: https://www.kaggle.com/code/vasant10/csiro-image2biomass-multi-output-regression", "3320938": "Does the train/test images have the same resolution?", "3311506": "i am having a issue with the submission , when i am running the notebook it is generating model weights and submission.csv in the same output folder but when i am submitting the notebook , i am getting the response that notebook is running without interrupting but \nsubmission.csv is not there ", "3311447": "@qiyuliaocsiro please correct me.\n\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g, \nGDM_g = Dry_Total_g - Dry_Dead_g\n\nso we have to predict Dry_Green_g , Dry_Clover_g , Dry_Dead_g.\n\n\"Dry_Dead_g\" is the biomass(clover + other_plants) which are completely dead and dried.\n\n\"Dry_Green_g\" is the biomass(other_plants) which are dried but alive.\n\n\"Dry_Clover_g\" is the biomass(clover) which are dried but alive.", "3311340": "This is my first competition, i am having difficulty uploading my file. It passes but it says submission failed. anyone know what i could be missing?", "3311044": "hello, can we use metadata like NDVI or other fileds during training??", "3310894": "Do I need to submit a notebook along with the prediction.csv? The 'Submit Prediction' button does not work. Is there an alternative way of submitting predictions? Patiently waiting for a response.", "3310649": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F23193939%2Fca95e1555a93fa3d256c37770cb91859%2FID670276799.jpg?generation=1762165289757598&alt=media)\nI’d like to ask about this image — what’s going on with it? It still has a watermark.\nAlso, do we need to take the climate into consideration?", "3309797": "Is the test data missing something? When I downloaded it, it contained only 5-6 records and 1 image without any numerical features. ", "3309057": "Hello , I am wondering how can I have the train dataset. Any advice on this?", "3308983": "Regarding the Sampling_Date variable, is it advisable to use cyclic coding (sin/cos) to represent the month and day of the year in order to capture seasonal patterns in biomass, or is there another date feature extraction technique that is considered more appropriate in this context?", "3308860": "Could you please provide the size of the actual test dataset to calculate the inference time?", "3308856": "Hello , how to get the dataset please ", "3308842": "hello, can we use metadata like NDVI or other fileds during training??", "3308647": "Sorry. I am not able to submit the submission.csv. Can someone help me with it?", "3308488": "Hello, \n\ni have a question regarding: Dry green vegetation (excluding clover)\nShould we filter images that have \"Species\" == \"clover?\nIf yes what value should the predicted Dry green vegetation have?\n\nthank you in advance", "3309944": "" } }