Hello Kagglers! I’m having trouble using the Sampling_Date, State, Species, Pre_GSHH_NDVI, Height_Ave_cm, and target_name columns for inference, but the test data only contains images. How can these columns be utilized? Thanks for your response!
", "messages": [ { "id": 3322093, "postDate": "2025-11-13T10:09:36.960Z", "content": "Hello Kagglers! I’m having trouble using the Sampling_Date, State, Species, Pre_GSHH_NDVI, Height_Ave_cm, and target_name columns for inference, but the test data only contains images. How can these columns be utilized? Thanks for your response!
", "rawMarkdown": "Hello Kagglers! I’m having trouble using the Sampling_Date, State, Species, Pre_GSHH_NDVI, Height_Ave_cm, and target_name columns for inference, but the test data only contains images. How can these columns be utilized? Thanks for your response!", "votes": 8 }, { "id": 3323748, "postDate": "2025-11-14T12:51:33.380Z", "content": "It should be possible to calculate NDVi but the required image band is not available (Near Infrared). Some other indices can be calculated using formular (RGB). But I tried already and it dropped my r2
", "rawMarkdown": "It should be possible to calculate NDVi but the required image band is not available (Near Infrared). Some other indices can be calculated using formular (RGB). But I tried already and it dropped my r2", "votes": 1 }, { "id": 3322436, "postDate": "2025-11-13T14:17:33.083Z", "content": "Only way to use them is to build a model that predicts those from the training images, than predict the values on test. Model to predict the meta data that I built does pretty good job of predicting the metadata but it did not work well (poor cv) when attempting to do the whole thing on local machine. Not sure if I had errors or if its just a bad idea.
", "rawMarkdown": "Only way to use them is to build a model that predicts those from the training images, than predict the values on test. Model to predict the meta data that I built does pretty good job of predicting the metadata but it did not work well (poor cv) when attempting to do the whole thing on local machine. Not sure if I had errors or if its just a bad idea.", "votes": 1, "replies": [ { "id": 3322747, "postDate": "2025-11-13T17:20:47.650Z", "content": "I tried a two-stage model: first predicting the metadata, then using the metadata to predict the target values. However, it wasn’t effective, and I ended up with a very poor LB score.
", "rawMarkdown": "I tried a two-stage model: first predicting the metadata, then using the metadata to predict the target values. However, it wasn’t effective, and I ended up with a very poor LB score.", "votes": 2 }, { "id": 3323224, "postDate": "2025-11-14T04:21:27.680Z", "content": "did the same, LB was 0.36 😅
", "rawMarkdown": "did the same, LB was 0.36 😅", "votes": 1, "replies": [ { "id": 3323382, "postDate": "2025-11-14T06:56:38.443Z", "content": "Yes, some results got negative 😅
", "rawMarkdown": "Yes, some results got negative 😅" } ] }, { "id": 3323503, "postDate": "2025-11-14T08:54:25.950Z", "content": "I tried the same. But I believe predicting the height and nvdi are actually very hard using only images. Then, I also tried detaching the backbone features before passing them to the height and nvdi heads so that they don't ruin the weights of the backbone. But that didn't help either.
", "rawMarkdown": "I tried the same. But I believe predicting the height and nvdi are actually very hard using only images. Then, I also tried detaching the backbone features before passing them to the height and nvdi heads so that they don't ruin the weights of the backbone. But that didn't help either.", "replies": [ { "id": 3323925, "postDate": "2025-11-14T15:04:22.710Z", "content": "身高要 用草的先後順序根草的養角來判別至於仰角能用草的生長點跟影子判定
", "rawMarkdown": "身高要 用草的先後順序根草的養角來判別至於仰角能用草的生長點跟影子判定\n\n\n", "votes": -5 } ] } ] }, { "id": 3339612, "postDate": "2025-11-19T06:30:04.737Z", "content": "You could try to adapt a two stage approach, which would be great, however, you have to keep in mind, that those values i.e. Metadata even though taken many times have still a small amount of error in them. Example the NDVI from my understanding depends also on the daytime you take a measurement, the density and also the species. I think the purpose of this challenge is to come up with an Idea how ** to not use those** Meta Data because they are very costly and not accessible to every farmer.
", "rawMarkdown": "You could try to adapt a two stage approach, which would be great, however, you have to keep in mind, that those values i.e. Metadata even though taken many times have still a small amount of error in them. Example the NDVI from my understanding depends also on the daytime you take a measurement, the density and also the species. I think the purpose of this challenge is to come up with an Idea how ** to not use those** Meta Data because they are very costly and not accessible to every farmer. " }, { "id": 3322588, "postDate": "2025-11-13T15:38:46.173Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 3322737, "postDate": "2025-11-13T17:10:27.997Z", "content": "Thanks! I built a model with an accuracy of 0.59 on the leaderboard, but further features adding didn’t help. I then tried to apply metadata, but ran into the problem that the hidden test didn’t include it.
", "rawMarkdown": "Thanks! I built a model with an accuracy of 0.59 on the leaderboard, but further features adding didn’t help. I then tried to apply metadata, but ran into the problem that the hidden test didn’t include it.", "replies": [ { "id": 3322989, "postDate": "2025-11-13T23:47:29.017Z", "content": "Hi Rustam, metadata is not provided in the testing set.
", "rawMarkdown": "Hi Rustam, metadata is not provided in the testing set." }, { "id": 3323065, "postDate": "2025-11-14T01:31:52.520Z", "content": "Thank you clarifiying!
", "rawMarkdown": "Thank you clarifiying!" }, { "id": 3329359, "postDate": "2025-11-16T05:19:47.870Z", "content": "Is there a reason for not including it in the test set?
", "rawMarkdown": "Is there a reason for not including it in the test set?" }, { "id": 3330124, "postDate": "2025-11-16T07:38:54.033Z", "content": "Maybe they just want to take a photo and predict the target values.
", "rawMarkdown": "Maybe they just want to take a photo and predict the target values." } ] } ] } ], "comments": [ { "id": 3323748, "author_name": "William Alabi", "author_url": "", "post_date": "2025-11-14T12:51:33.380000", "content": "It should be possible to calculate NDVi but the required image band is not available (Near Infrared). Some other indices can be calculated using formular (RGB). But I tried already and it dropped my r2
", "votes": 1, "replies": [] }, { "id": 3322436, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-11-13T14:17:33.083000", "content": "Only way to use them is to build a model that predicts those from the training images, than predict the values on test. Model to predict the meta data that I built does pretty good job of predicting the metadata but it did not work well (poor cv) when attempting to do the whole thing on local machine. Not sure if I had errors or if its just a bad idea.
", "votes": 1, "replies": [ { "id": 3322747, "author_name": "Rustam Bazarbayev", "author_url": "", "post_date": "2025-11-13T17:20:47.650000", "content": "I tried a two-stage model: first predicting the metadata, then using the metadata to predict the target values. However, it wasn’t effective, and I ended up with a very poor LB score.
", "votes": 2, "replies": [] }, { "id": 3323224, "author_name": "Geremie Yeo", "author_url": "", "post_date": "2025-11-14T04:21:27.680000", "content": "did the same, LB was 0.36 😅
", "votes": 1, "replies": [ { "id": 3323382, "author_name": "Rustam Bazarbayev", "author_url": "", "post_date": "2025-11-14T06:56:38.443000", "content": "Yes, some results got negative 😅
", "votes": 0, "replies": [] } ] }, { "id": 3323503, "author_name": "Komil Parmar", "author_url": "", "post_date": "2025-11-14T08:54:25.950000", "content": "I tried the same. But I believe predicting the height and nvdi are actually very hard using only images. Then, I also tried detaching the backbone features before passing them to the height and nvdi heads so that they don't ruin the weights of the backbone. But that didn't help either.
", "votes": 0, "replies": [ { "id": 3323925, "author_name": "dddff54", "author_url": "", "post_date": "2025-11-14T15:04:22.710000", "content": "身高要 用草的先後順序根草的養角來判別至於仰角能用草的生長點跟影子判定
", "votes": -5, "replies": [] } ] } ] }, { "id": 3339612, "author_name": "Don", "author_url": "", "post_date": "2025-11-19T06:30:04.737000", "content": "You could try to adapt a two stage approach, which would be great, however, you have to keep in mind, that those values i.e. Metadata even though taken many times have still a small amount of error in them. Example the NDVI from my understanding depends also on the daytime you take a measurement, the density and also the species. I think the purpose of this challenge is to come up with an Idea how ** to not use those** Meta Data because they are very costly and not accessible to every farmer.
", "votes": 0, "replies": [] }, { "id": 3322588, "author_name": "", "author_url": "", "post_date": "2025-11-13T15:38:46.173000", "content": "", "votes": 0, "replies": [ { "id": 3322737, "author_name": "Rustam Bazarbayev", "author_url": "", "post_date": "2025-11-13T17:10:27.997000", "content": "Thanks! I built a model with an accuracy of 0.59 on the leaderboard, but further features adding didn’t help. I then tried to apply metadata, but ran into the problem that the hidden test didn’t include it.
", "votes": 0, "replies": [ { "id": 3322989, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-11-13T23:47:29.017000", "content": "Hi Rustam, metadata is not provided in the testing set.
", "votes": 0, "replies": [] }, { "id": 3323065, "author_name": "Rustam Bazarbayev", "author_url": "", "post_date": "2025-11-14T01:31:52.520000", "content": "Thank you clarifiying!
", "votes": 0, "replies": [] }, { "id": 3329359, "author_name": "Joseph Roccisano", "author_url": "", "post_date": "2025-11-16T05:19:47.870000", "content": "Is there a reason for not including it in the test set?
", "votes": 0, "replies": [] }, { "id": 3330124, "author_name": "Rustam Bazarbayev", "author_url": "", "post_date": "2025-11-16T07:38:54.033000", "content": "Maybe they just want to take a photo and predict the target values.
", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3322093": "Hello Kagglers! I’m having trouble using the Sampling_Date, State, Species, Pre_GSHH_NDVI, Height_Ave_cm, and target_name columns for inference, but the test data only contains images. How can these columns be utilized? Thanks for your response!", "3323748": "It should be possible to calculate NDVi but the required image band is not available (Near Infrared). Some other indices can be calculated using formular (RGB). But I tried already and it dropped my r2", "3322436": "Only way to use them is to build a model that predicts those from the training images, than predict the values on test. Model to predict the meta data that I built does pretty good job of predicting the metadata but it did not work well (poor cv) when attempting to do the whole thing on local machine. Not sure if I had errors or if its just a bad idea.", "3339612": "You could try to adapt a two stage approach, which would be great, however, you have to keep in mind, that those values i.e. Metadata even though taken many times have still a small amount of error in them. Example the NDVI from my understanding depends also on the daytime you take a measurement, the density and also the species. I think the purpose of this challenge is to come up with an Idea how ** to not use those** Meta Data because they are very costly and not accessible to every farmer. ", "3322588": "" } }