![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2956018%2F2bea030f15767ccf74b8a8ea2ac11cb3%2Finbox_2956018_1c4847359843c731116a439d18ff8f26_.png?generation=1564888464243557&alt=media\")
\nAnd it seems that many people face the same peoblem as this. So is this Kaggle's kernel bugs?We're very excited to launch this experimental competition with you. It's a first in many ways - running neural nets in the evaluation pipeline, accepting images as submissions instead of csvs, etc. We hope you enjoy this competition and learn a lot from it.
\n\nPlease report any technical issues or feedback in this thread.
", "messages": [ { "id": "563826", "postDate": "06/28/2019 18:08:28", "content": "We're very excited to launch this experimental competition with you. It's a first in many ways - running neural nets in the evaluation pipeline, accepting images as submissions instead of csvs, etc. We hope you enjoy this competition and learn a lot from it.
\n\nPlease report any technical issues or feedback in this thread.
", "rawMarkdown": "We're very excited to launch this experimental competition with you. It's a first in many ways - running neural nets in the evaluation pipeline, accepting images as submissions instead of csvs, etc. We hope you enjoy this competition and learn a lot from it. \n\nPlease report any technical issues or feedback in this thread.", "votes": null }, { "id": "563847", "postDate": "06/28/2019 18:29:17", "content": "Thank you for hosting! \nI'm excited to join this interesting competition.
", "rawMarkdown": "Thank you for hosting! \nI'm excited to join this interesting competition.", "votes": null }, { "id": "563870", "postDate": "06/28/2019 18:57:06", "content": "Is it allowed to submit 10,000 times the same image (or a slight variation of it)? If not, is it in the rules and how can you prevent this?
", "rawMarkdown": "Is it allowed to submit 10,000 times the same image (or a slight variation of it)? If not, is it in the rules and how can you prevent this?", "votes": null }, { "id": "563902", "postDate": "06/28/2019 19:11:24", "content": "FID (and MiFID) will be quite bad if that is the case.
", "rawMarkdown": "FID (and MiFID) will be quite bad if that is the case.", "votes": null }, { "id": "563938", "postDate": "06/28/2019 19:41:15", "content": "Thanks, will check the metric out.
", "rawMarkdown": "Thanks, will check the metric out.", "votes": null }, { "id": "564024", "postDate": "06/28/2019 22:55:27", "content": "I have two questions regarding the competition rules:
\n\nAre imagenet pre-trained models allowed?
All the training/inference must be made in the same kernel or, as in other kernel competitions, we can train offline and upload a dataset with the weights?
Thanks!
", "rawMarkdown": "I have two questions regarding the competition rules:\n\n- Are imagenet pre-trained models allowed? \n\n- All the training/inference must be made in the same kernel or, as in other kernel competitions, we can train offline and upload a dataset with the weights?\n\nThanks!", "votes": null }, { "id": "564046", "postDate": "06/28/2019 23:32:05", "content": "The answer is No and No.
\n\nAs stated in our Kernels requirement page:
\n\n\n\n\n\n
\n- No internet access enabled
\n- No external data can be added as a data source
\n- No use of other kernels (or utility scripts) as inputs into your submission kernel
\n
Furthermore, if you add other Datasets in your Kernel, you will not be able to submit those into the competition (the submit to competition button grays out)
Great competition, but Kaggle should make clarifications on some issues, for example, whether the following blend method could be used?\nhttps://www.kaggle.com/jazivxt/imitation-game/notebook
", "rawMarkdown": "Great competition, but Kaggle should make clarifications on some issues, for example, whether the following blend method could be used?\nhttps://www.kaggle.com/jazivxt/imitation-game/notebook", "votes": null }, { "id": "564334", "postDate": "06/29/2019 09:38:18", "content": "Hi @wendykan , I have 2 questions:
\n\n@wendykan could you clarify wich generative models are you looking for? only GANs?
\n\n\n\n\nCompetitors must use generative methods to create their submission images and are not permitted to make submissions that include any images already classified as dogs or altered versions of such images....
\n
Moreover, are not permitted to make submissions that include ... altered versions of such images many generative methods work like this. So, What does kaggle expect?
\n\n\nUse your training skills to create images, rather than identify them. You’ll be using GANs, which are at the creative frontier of machine learning....
\n
Thanks.
", "rawMarkdown": "wendykan could you clarify wich generative models are you looking for? only GANs?\n\n> Competitors **must use generative methods** to create their submission images and are not permitted to make submissions that include any images already classified as dogs or altered versions of such images....\n\nMoreover, ```are not permitted to make submissions that include ... altered versions of such images``` many generative methods work like this. So, What does kaggle expect?\n\n---\n\n> Use your training skills to create images, rather than identify them. **You’ll be using GANs**, which are at the creative frontier of machine learning....\n\nThanks.", "votes": null }, { "id": "566036", "postDate": "07/01/2019 18:20:46", "content": "Many thanks for giving even more possibilities and diversity to competitions, im enjoying a lot this competition :D
", "rawMarkdown": "Many thanks for giving even more possibilities and diversity to competitions, im enjoying a lot this competition :D", "votes": null }, { "id": "567372", "postDate": "07/03/2019 12:14:54", "content": "@wendykan Could you please clarify about using external data? Because in the Kernels requirement page we have this sentence:
\n\n\n\n\nNo external data can be added as a data source
\n
But in the rules page it says:
\n\n\n\n\nC. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your models and Submissions. However, you will (i) ensure the External Data is available to use by all participants of the competition for purposes of the competition at no cost to the other participants and (ii) post such access to the External Data for the participants to the official competition forum prior to the Entry Deadline.
\n
Please which one is correct?
", "rawMarkdown": "wendykan Could you please clarify about using external data? Because in the [Kernels requirement](https://www.kaggle.com/c/generative-dog-images/overview/kernels-requirements) page we have this sentence:\n> No external data can be added as a data source\n\nBut in the [rules](https://www.kaggle.com/c/generative-dog-images/rules) page it says:\n> C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your models and Submissions. However, you will (i) ensure the External Data is available to use by all participants of the competition for purposes of the competition at no cost to the other participants and (ii) post such access to the External Data for the participants to the official competition forum prior to the Entry Deadline.\n\nPlease which one is correct?", "votes": null }, { "id": "567473", "postDate": "07/03/2019 15:23:11", "content": "Apologies for the confusion. The Competition specific rules in section A supersede the General rules. So external data is not permitted, and is systematically prohibited in kernels, in that, if you add external data sources/utility scripts or turn on internet access, you will not be able to submit your kernel to the competition.
", "rawMarkdown": "Apologies for the confusion. The Competition specific rules in section A supersede the General rules. So external data is not permitted, and is systematically prohibited in kernels, in that, if you add external data sources/utility scripts or turn on internet access, you will not be able to submit your kernel to the competition.", "votes": null }, { "id": "567485", "postDate": "07/03/2019 15:30:14", "content": "Ok thank you!
", "rawMarkdown": "Ok thank you!", "votes": null }, { "id": "568598", "postDate": "07/05/2019 07:21:56", "content": "Must we use GAN? Could we use VAE or any other generative models?
", "rawMarkdown": "Must we use GAN? Could we use VAE or any other generative models?", "votes": null }, { "id": "571557", "postDate": "07/09/2019 18:19:03", "content": "", "rawMarkdown": "[See answer here](https://www.kaggle.com/c/generative-dog-images/discussion/98183#latest-571434)", "votes": null }, { "id": "577484", "postDate": "07/16/2019 18:11:27", "content": "Hi, @wendykan, I have a question about the evaluation.
\n\nAs I understand from the graph in the evaluation page, when user commits his/her kernel, it only generates the dog images. When user submits to competition, there's no kernel re-running again, but only calculating the score with the generated dog images. Is this correct?
\n\nIf the above is correct, during scoring (after submitting to competition), for private score, the cosine distance calculated is between features from the generated dog images and the mysterious private dog images, which the user kernel never sees before generating dog images, then what's the point to check distance between them? I mean user kernel can never memorizes those private dog images in any cases because it never sees them. And if accidentally the distance is close, then the evaluation is bad while the image is totally legit to be good. Am I right?
\n\nIf the kernel is indeed rerunning after submitting to competition, is it using private dog images, instead of public dog images, for training? This would be not the case shown in your graph in the evaluation page. It's better to clarify.
", "rawMarkdown": "Hi, @wendykan, I have a question about the evaluation.\n\nAs I understand from the graph in the evaluation page, when user commits his/her kernel, it only generates the dog images. When user submits to competition, there's no kernel re-running again, but only calculating the score with the generated dog images. Is this correct?\n\nIf the above is correct, during scoring (after submitting to competition), for private score, the cosine distance calculated is between features from the generated dog images and the mysterious private dog images, which the user kernel never sees before generating dog images, then what's the point to check distance between them? I mean user kernel can never memorizes those private dog images in any cases because it never sees them. And if accidentally the distance is close, then the evaluation is bad while the image is totally legit to be good. Am I right?\n\nIf the kernel is indeed rerunning after submitting to competition, is it using private dog images, instead of public dog images, for training? This would be not the case shown in your graph in the evaluation page. It's better to clarify.", "votes": null }, { "id": "591618", "postDate": "08/04/2019 03:17:22", "content": "Hi, @wendykan , I got a kernel exception with the unknown error on submission.\n\nAnd it seems that many people face the same peoblem as this. So is this Kaggle's kernel bugs?
Thank you for hosting! \nI'm excited to join this interesting competition.
", "votes": null, "replies": [] }, { "id": 563870, "author_name": "philippsinger", "author_url": "", "post_date": "06/28/2019 18:57:06", "content": "Is it allowed to submit 10,000 times the same image (or a slight variation of it)? If not, is it in the rules and how can you prevent this?
", "votes": null, "replies": [ { "id": 563902, "author_name": "wendykan", "author_url": "", "post_date": "06/28/2019 19:11:24", "content": "FID (and MiFID) will be quite bad if that is the case.
", "votes": null, "replies": [] }, { "id": 563938, "author_name": "philippsinger", "author_url": "", "post_date": "06/28/2019 19:41:15", "content": "Thanks, will check the metric out.
", "votes": null, "replies": [] } ] }, { "id": 564024, "author_name": "mnpinto", "author_url": "", "post_date": "06/28/2019 22:55:27", "content": "I have two questions regarding the competition rules:
\n\nAre imagenet pre-trained models allowed?
All the training/inference must be made in the same kernel or, as in other kernel competitions, we can train offline and upload a dataset with the weights?
Thanks!
", "votes": null, "replies": [ { "id": 564046, "author_name": "wendykan", "author_url": "", "post_date": "06/28/2019 23:32:05", "content": "The answer is No and No.
\n\nAs stated in our Kernels requirement page:
\n\n\n\n\n\n
\n- No internet access enabled
\n- No external data can be added as a data source
\n- No use of other kernels (or utility scripts) as inputs into your submission kernel
\n
Furthermore, if you add other Datasets in your Kernel, you will not be able to submit those into the competition (the submit to competition button grays out)
Great competition, but Kaggle should make clarifications on some issues, for example, whether the following blend method could be used?\nhttps://www.kaggle.com/jazivxt/imitation-game/notebook
", "votes": null, "replies": [] }, { "id": 564334, "author_name": "jesucristo", "author_url": "", "post_date": "06/29/2019 09:38:18", "content": "Hi @wendykan , I have 2 questions:
\n\n@wendykan could you clarify wich generative models are you looking for? only GANs?
\n\n\n\n\nCompetitors must use generative methods to create their submission images and are not permitted to make submissions that include any images already classified as dogs or altered versions of such images....
\n
Moreover, are not permitted to make submissions that include ... altered versions of such images many generative methods work like this. So, What does kaggle expect?
\n\n\nUse your training skills to create images, rather than identify them. You’ll be using GANs, which are at the creative frontier of machine learning....
\n
Thanks.
", "votes": null, "replies": [] }, { "id": 566036, "author_name": "enric1296", "author_url": "", "post_date": "07/01/2019 18:20:46", "content": "Many thanks for giving even more possibilities and diversity to competitions, im enjoying a lot this competition :D
", "votes": null, "replies": [] }, { "id": 567372, "author_name": "felipefonte99", "author_url": "", "post_date": "07/03/2019 12:14:54", "content": "@wendykan Could you please clarify about using external data? Because in the Kernels requirement page we have this sentence:
\n\n\n\n\nNo external data can be added as a data source
\n
But in the rules page it says:
\n\n\n\n\nC. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your models and Submissions. However, you will (i) ensure the External Data is available to use by all participants of the competition for purposes of the competition at no cost to the other participants and (ii) post such access to the External Data for the participants to the official competition forum prior to the Entry Deadline.
\n
Please which one is correct?
", "votes": null, "replies": [ { "id": 567473, "author_name": "juliaelliott", "author_url": "", "post_date": "07/03/2019 15:23:11", "content": "Apologies for the confusion. The Competition specific rules in section A supersede the General rules. So external data is not permitted, and is systematically prohibited in kernels, in that, if you add external data sources/utility scripts or turn on internet access, you will not be able to submit your kernel to the competition.
", "votes": null, "replies": [] }, { "id": 567485, "author_name": "felipefonte99", "author_url": "", "post_date": "07/03/2019 15:30:14", "content": "Ok thank you!
", "votes": null, "replies": [] } ] }, { "id": 568598, "author_name": "seefun", "author_url": "", "post_date": "07/05/2019 07:21:56", "content": "Must we use GAN? Could we use VAE or any other generative models?
", "votes": null, "replies": [ { "id": 571557, "author_name": "juliaelliott", "author_url": "", "post_date": "07/09/2019 18:19:03", "content": "", "votes": null, "replies": [] } ] }, { "id": 577484, "author_name": "amigd23", "author_url": "", "post_date": "07/16/2019 18:11:27", "content": "Hi, @wendykan, I have a question about the evaluation.
\n\nAs I understand from the graph in the evaluation page, when user commits his/her kernel, it only generates the dog images. When user submits to competition, there's no kernel re-running again, but only calculating the score with the generated dog images. Is this correct?
\n\nIf the above is correct, during scoring (after submitting to competition), for private score, the cosine distance calculated is between features from the generated dog images and the mysterious private dog images, which the user kernel never sees before generating dog images, then what's the point to check distance between them? I mean user kernel can never memorizes those private dog images in any cases because it never sees them. And if accidentally the distance is close, then the evaluation is bad while the image is totally legit to be good. Am I right?
\n\nIf the kernel is indeed rerunning after submitting to competition, is it using private dog images, instead of public dog images, for training? This would be not the case shown in your graph in the evaluation page. It's better to clarify.
", "votes": null, "replies": [] }, { "id": 591618, "author_name": "adiachen", "author_url": "", "post_date": "08/04/2019 03:17:22", "content": "Hi, @wendykan , I got a kernel exception with the unknown error on submission.\n\nAnd it seems that many people face the same peoblem as this. So is this Kaggle's kernel bugs?