![\"EfferalGAN\"](\"https://www.upsa-nosproduits.com/sites/default/files/styles/produit/public/3d_efferalgan_1000_berry_0.png\")
It took a day to hack the competition https://www.kaggle.com/jazivxt/imitation-game\nIt seems that the rest of the month we will spend on discussing what is legal and what is not...
\n\n\n", "messages": [ { "id": 564308, "postDate": "2019-06-29T09:05:25.883Z", "content": "Important Warning: This competition has an experimental format and submission style (images as submission). 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
It took a day to hack the competition https://www.kaggle.com/jazivxt/imitation-game\nIt seems that the rest of the month we will spend on discussing what is legal and what is not...
\n\n\n", "rawMarkdown": "It took a day to hack the competition https://www.kaggle.com/jazivxt/imitation-game\nIt seems that the rest of the month we will spend on discussing what is legal and what is not...\n\n> Important Warning: This competition has an experimental format and submission style (images as submission). 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\n", "votes": 30 }, { "id": 564414, "postDate": "2019-06-29T11:37:44.413Z", "content": "Important Warning: This competition has an experimental format and submission style (images as submission). 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
\n\n\nIt seems that the rest of the month we will spend on discussing what is legal and what is not…
\n
Prepare your EfferalGAN kaggle admins. You'll have some headaches ;)
\n\n
Hahahah, Rofl
", "rawMarkdown": "Hahahah, Rofl" } ] }, { "id": 564691, "postDate": "2019-06-29T20:26:13.190Z", "content": "This is a fun competition but it will be very difficult to define generative methods. Are we restricting ourselves to GANs? According to Wikipedia here, types of generative models are:
\n\nWhat is blending two training images? Is that a Gaussian Mixture Model where each training observation is in its own sub population?
", "rawMarkdown": "This is a fun competition but it will be very difficult to define **generative methods**. Are we restricting ourselves to GANs? According to Wikipedia [here][1], types of generative models are:\n\n* Gaussian mixture model (and other types of mixture model)\n* Hidden Markov model\n* Probabilistic context-free grammar\n* Bayesian network (e.g. Naive bayes, Autoregressive model)\n* Averaged one-dependence estimators\n* Latent Dirichlet allocation\n* Boltzmann machine (e.g. Restricted Boltzmann machine, Deep belief network)\n* Variational autoencoder\n* Generative adversarial network\n* Flow-based generative model\n\nWhat is blending two training images? Is that a Gaussian Mixture Model where each training observation is in its own sub population?\n\n[1]: https://en.wikipedia.org/wiki/Generative_model#Generative_models", "votes": 11, "replies": [ { "id": 564809, "postDate": "2019-06-30T03:01:11.053Z", "content": "Kaggle should fix the rules stating that everyone MUST use Generative Adversarial Networks in order to solve this problem. But at the same time its not easy to define a robust metric to evaluate a generative model.
\n\nHaving external data can be controlled disabling Kernel internet access. But people still can upload some dog images to private datasets to build submissions. Also there are other ways of cheating that don't involve a simple blend of true dog images.
\n\nMaybe disabling medals and ranking points is a better way to test this kind of competition.
", "rawMarkdown": "Kaggle should fix the rules stating that everyone MUST use Generative Adversarial Networks in order to solve this problem. But at the same time its not easy to define a robust metric to evaluate a generative model. \n\nHaving external data can be controlled disabling Kernel internet access. But people still can upload some dog images to private datasets to build submissions. Also there are other ways of cheating that don't involve a simple blend of true dog images. \n\nMaybe disabling medals and ranking points is a better way to test this kind of competition.", "votes": 10 }, { "id": 565321, "postDate": "2019-06-30T18:55:08.963Z", "content": "Especially on the blending images. StyleGAN allows you to \"blend\" the features of two images but it still remains to be a generative model(and a GAN). Now whether or not someone will laboriously train a StyleGAN is still a question
", "rawMarkdown": "Especially on the blending images. StyleGAN allows you to \"blend\" the features of two images but it still remains to be a generative model(and a GAN). Now whether or not someone will laboriously train a StyleGAN is still a question", "votes": 1 } ] }, { "id": 564409, "postDate": "2019-06-29T11:28:35.153Z", "content": "It's starting ...
\n\n", "rawMarkdown": "It's starting ...\n\n![\"Captura-de-pantalla-de-2019-06-29-13-27-06\"](\"https://i.ibb.co/MnSTLBT/Captura-de-pantalla-de-2019-06-29-13-27-06.png\")
",
"votes": 1
},
{
"id": 564323,
"postDate": "2019-06-29T09:26:31.690Z",
"content": "@jazivxt Thank you for the quick reaction.\nPlease make it private, or more people will use it and they can destroy the LB …
\n\nNow, @wendykan how do you check the images were generated using only generative methods? I mean, some people can use GAN as a \"smoke screen\", and then submit 1000 random images of dogs from Google
", "rawMarkdown": "@jazivxt Thank you for the quick reaction.\nPlease make it private, or more people will use it and they can destroy the LB …\n\nNow, @wendykan how do you check the images were generated using **only** generative methods? I mean, some people can use GAN as a \"smoke screen\", and then submit 1000 random images of dogs from Google ", "votes": 1, "replies": [ { "id": 564328, "postDate": "2019-06-29T09:29:21.523Z", "content": "there are countless ways to cheat here
", "rawMarkdown": "there are countless ways to cheat here", "votes": 3 }, { "id": 564330, "postDate": "2019-06-29T09:34:19.020Z", "content": "
", "rawMarkdown": "", "votes": 1 }, { "id": 564342, "postDate": "2019-06-29T09:42:31.297Z", "content": "This competition just cries for exploits and drama in the end.
", "rawMarkdown": "This competition just cries for exploits and drama in the end.", "votes": 14 }, { "id": 564611, "postDate": "2019-06-29T17:39:30.683Z", "content": "This is our best attempt to try something extremely new on Kaggle - we understand there are risks. We will do our best in checking the winning solutions. However, as you can imagine, if there's too much cheating, we will not be able to run these new and more cutting-edge competitions again in the future.
", "rawMarkdown": "This is our best attempt to try something extremely new on Kaggle - we understand there are risks. We will do our best in checking the winning solutions. However, as you can imagine, if there's too much cheating, we will not be able to run these new and more cutting-edge competitions again in the future. ", "votes": 8 }, { "id": 564618, "postDate": "2019-06-29T17:54:01.013Z", "content": "I do appreciate kaggle trying out such extremely new things and this competition in particular. I just expected it to have no prize fund / no ranking points granted as it is extremely prone to tricks and exploits which bring no added value to the end goal. Having e.g. prizes for top kernels or for finding exploits might have made more sense here.\nBut regardless, I hope this competition will end up great fun for participants and organizers!
", "rawMarkdown": "I do appreciate kaggle trying out such extremely new things and this competition in particular. I just expected it to have no prize fund / no ranking points granted as it is extremely prone to tricks and exploits which bring no added value to the end goal. Having e.g. prizes for top kernels or for finding exploits might have made more sense here.\nBut regardless, I hope this competition will end up great fun for participants and organizers!", "votes": 11 }, { "id": 564667, "postDate": "2019-06-29T18:58:48.210Z", "content": "agree, getting rid of ranking points would be a solution
", "rawMarkdown": "agree, getting rid of ranking points would be a solution" }, { "id": 564668, "postDate": "2019-06-29T19:12:25.407Z", "content": "Ｈ𝐀𝑷𝑷𝓎 🇰𝗮𝘨𝘨🇱𝖎Ｎɢ 💯
", "rawMarkdown": "Ｈ𝐀𝑷𝑷𝓎 🇰𝗮𝘨𝘨🇱𝖎Ｎɢ 💯", "votes": 3 }, { "id": 564672, "postDate": "2019-06-29T19:28:31.347Z", "content": "@jazivxt I belive that finding and pointing out tricks/exploits early and publicly, the way you did it, should be appreciated and rewarded. Even if it might not be of a benefit for the competition task, I hope it can be valuable for kaggle to take into account, react, improve and introduce new cutting-edge competitions.
", "rawMarkdown": "@jazivxt I belive that finding and pointing out tricks/exploits early and publicly, the way you did it, should be appreciated and rewarded. Even if it might not be of a benefit for the competition task, I hope it can be valuable for kaggle to take into account, react, improve and introduce new cutting-edge competitions.", "votes": 4 }, { "id": 564688, "postDate": "2019-06-29T20:13:53.670Z", "content": "totally agree with you. It is in a way like GANs themselves. We try to beat the metrics with non-GAN generated images, and they try to make better metrics to detect these. After a few cycles, the metric will be fine-tuned.
", "rawMarkdown": "totally agree with you. It is in a way like GANs themselves. We try to beat the metrics with non-GAN generated images, and they try to make better metrics to detect these. After a few cycles, the metric will be fine-tuned.", "votes": 2 } ] }, { "id": 564316, "postDate": "2019-06-29T09:16:22.560Z", "content": "Ｈ𝐀𝑷𝑷𝓎 🇰𝗮𝘨𝘨🇱𝖎Ｎɢ 💯
", "rawMarkdown": "Ｈ𝐀𝑷𝑷𝓎 🇰𝗮𝘨𝘨🇱𝖎Ｎɢ 💯", "votes": 1, "replies": [ { "id": 564319, "postDate": "2019-06-29T09:19:29.313Z", "content": "the problem with \"visual inspection\" is that it is impossible to check the entire LB, only first 5 (maybe 10) places will be checked
", "rawMarkdown": "the problem with \"visual inspection\" is that it is impossible to check the entire LB, only first 5 (maybe 10) places will be checked ", "votes": 2 }, { "id": 564333, "postDate": "2019-06-29T09:36:39.963Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 564839, "postDate": "2019-06-30T04:25:27.847Z", "content": "\n\n\nIt seems that the rest of the month we will spend on discussing what is legal and what is not…
\n
OK, one competition I won't enter. Arguing with other kagglers isn't fun. This one will be ugly I'm afraid.
\n\nI agree with @titericz , they should have imposed the use of GAN, or make it a swag only competition.
", "rawMarkdown": "> It seems that the rest of the month we will spend on discussing what is legal and what is not…\n\nOK, one competition I won't enter. Arguing with other kagglers isn't fun. This one will be ugly I'm afraid.\n\nI agree with @titericz , they should have imposed the use of GAN, or make it a swag only competition.", "votes": 1 } ], "comments": [ { "id": 564414, "author_name": "Serigne ", "author_url": "", "post_date": "2019-06-29T11:37:44.413000", "content": "\n\n\nIt seems that the rest of the month we will spend on discussing what is legal and what is not…
\n
Prepare your EfferalGAN kaggle admins. You'll have some headaches ;)
\n\n
Hahahah, Rofl
", "votes": 0, "replies": [] } ] }, { "id": 564691, "author_name": "Chris Deotte", "author_url": "", "post_date": "2019-06-29T20:26:13.190000", "content": "This is a fun competition but it will be very difficult to define generative methods. Are we restricting ourselves to GANs? According to Wikipedia here, types of generative models are:
\n\nWhat is blending two training images? Is that a Gaussian Mixture Model where each training observation is in its own sub population?
", "votes": 11, "replies": [ { "id": 564809, "author_name": "Giba", "author_url": "", "post_date": "2019-06-30T03:01:11.053000", "content": "Kaggle should fix the rules stating that everyone MUST use Generative Adversarial Networks in order to solve this problem. But at the same time its not easy to define a robust metric to evaluate a generative model.
\n\nHaving external data can be controlled disabling Kernel internet access. But people still can upload some dog images to private datasets to build submissions. Also there are other ways of cheating that don't involve a simple blend of true dog images.
\n\nMaybe disabling medals and ranking points is a better way to test this kind of competition.
", "votes": 10, "replies": [] }, { "id": 565321, "author_name": "sh", "author_url": "", "post_date": "2019-06-30T18:55:08.963000", "content": "Especially on the blending images. StyleGAN allows you to \"blend\" the features of two images but it still remains to be a generative model(and a GAN). Now whether or not someone will laboriously train a StyleGAN is still a question
", "votes": 1, "replies": [] } ] }, { "id": 564409, "author_name": "Nanashi", "author_url": "", "post_date": "2019-06-29T11:28:35.153000", "content": "It's starting ...
\n\n", "votes": 1, "replies": [] }, { "id": 564323, "author_name": "Nanashi", "author_url": "", "post_date": "2019-06-29T09:26:31.690000", "content": "@jazivxt Thank you for the quick reaction.\nPlease make it private, or more people will use it and they can destroy the LB …
\n\nNow, @wendykan how do you check the images were generated using only generative methods? I mean, some people can use GAN as a \"smoke screen\", and then submit 1000 random images of dogs from Google
", "votes": 1, "replies": [ { "id": 564328, "author_name": "", "author_url": "", "post_date": "2019-06-29T09:29:21.523000", "content": "there are countless ways to cheat here
", "votes": 3, "replies": [] }, { "id": 564330, "author_name": "Nanashi", "author_url": "", "post_date": "2019-06-29T09:34:19.020000", "content": "
", "votes": 1, "replies": [] }, { "id": 564342, "author_name": "Psi", "author_url": "", "post_date": "2019-06-29T09:42:31.297000", "content": "This competition just cries for exploits and drama in the end.
", "votes": 14, "replies": [] }, { "id": 564611, "author_name": "Wendy Kan", "author_url": "", "post_date": "2019-06-29T17:39:30.683000", "content": "This is our best attempt to try something extremely new on Kaggle - we understand there are risks. We will do our best in checking the winning solutions. However, as you can imagine, if there's too much cheating, we will not be able to run these new and more cutting-edge competitions again in the future.
", "votes": 8, "replies": [] }, { "id": 564618, "author_name": "dott", "author_url": "", "post_date": "2019-06-29T17:54:01.013000", "content": "I do appreciate kaggle trying out such extremely new things and this competition in particular. I just expected it to have no prize fund / no ranking points granted as it is extremely prone to tricks and exploits which bring no added value to the end goal. Having e.g. prizes for top kernels or for finding exploits might have made more sense here.\nBut regardless, I hope this competition will end up great fun for participants and organizers!
", "votes": 11, "replies": [] }, { "id": 564667, "author_name": "", "author_url": "", "post_date": "2019-06-29T18:58:48.210000", "content": "agree, getting rid of ranking points would be a solution
", "votes": 0, "replies": [] }, { "id": 564668, "author_name": "jazivxt", "author_url": "", "post_date": "2019-06-29T19:12:25.407000", "content": "Ｈ𝐀𝑷𝑷𝓎 🇰𝗮𝘨𝘨🇱𝖎Ｎɢ 💯
", "votes": 3, "replies": [] }, { "id": 564672, "author_name": "dott", "author_url": "", "post_date": "2019-06-29T19:28:31.347000", "content": "@jazivxt I belive that finding and pointing out tricks/exploits early and publicly, the way you did it, should be appreciated and rewarded. Even if it might not be of a benefit for the competition task, I hope it can be valuable for kaggle to take into account, react, improve and introduce new cutting-edge competitions.
", "votes": 4, "replies": [] }, { "id": 564688, "author_name": "Ehsan", "author_url": "", "post_date": "2019-06-29T20:13:53.670000", "content": "totally agree with you. It is in a way like GANs themselves. We try to beat the metrics with non-GAN generated images, and they try to make better metrics to detect these. After a few cycles, the metric will be fine-tuned.
", "votes": 2, "replies": [] } ] }, { "id": 564316, "author_name": "jazivxt", "author_url": "", "post_date": "2019-06-29T09:16:22.560000", "content": "Ｈ𝐀𝑷𝑷𝓎 🇰𝗮𝘨𝘨🇱𝖎Ｎɢ 💯
", "votes": 1, "replies": [ { "id": 564319, "author_name": "", "author_url": "", "post_date": "2019-06-29T09:19:29.313000", "content": "the problem with \"visual inspection\" is that it is impossible to check the entire LB, only first 5 (maybe 10) places will be checked
", "votes": 2, "replies": [] }, { "id": 564333, "author_name": "", "author_url": "", "post_date": "2019-06-29T09:36:39.963000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 564839, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-30T04:25:27.847000", "content": "\n\n\nIt seems that the rest of the month we will spend on discussing what is legal and what is not…
\n
OK, one competition I won't enter. Arguing with other kagglers isn't fun. This one will be ugly I'm afraid.
\n\nI agree with @titericz , they should have imposed the use of GAN, or make it a swag only competition.
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "564308": "It took a day to hack the competition https://www.kaggle.com/jazivxt/imitation-game\nIt seems that the rest of the month we will spend on discussing what is legal and what is not...\n\n> Important Warning: This competition has an experimental format and submission style (images as submission). 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\n", "564414": ">It seems that the rest of the month we will spend on discussing what is legal and what is not…\n\nPrepare your EfferalGAN kaggle admins. You'll have some headaches ;) \n\n", "564691": "This is a fun competition but it will be very difficult to define **generative methods**. Are we restricting ourselves to GANs? According to Wikipedia [here][1], types of generative models are:\n\n* Gaussian mixture model (and other types of mixture model)\n* Hidden Markov model\n* Probabilistic context-free grammar\n* Bayesian network (e.g. Naive bayes, Autoregressive model)\n* Averaged one-dependence estimators\n* Latent Dirichlet allocation\n* Boltzmann machine (e.g. Restricted Boltzmann machine, Deep belief network)\n* Variational autoencoder\n* Generative adversarial network\n* Flow-based generative model\n\nWhat is blending two training images? Is that a Gaussian Mixture Model where each training observation is in its own sub population?\n\n[1]: https://en.wikipedia.org/wiki/Generative_model#Generative_models", "564409": "It's starting ...\n\n",
"564323": "@jazivxt Thank you for the quick reaction.\nPlease make it private, or more people will use it and they can destroy the LB …\n\nNow, @wendykan how do you check the images were generated using **only** generative methods? I mean, some people can use GAN as a \"smoke screen\", and then submit 1000 random images of dogs from Google ",
"564316": "Ｈ𝐀𝑷𝑷𝓎 🇰𝗮𝘨𝘨🇱𝖎Ｎɢ 💯",
"564839": "> It seems that the rest of the month we will spend on discussing what is legal and what is not…\n\nOK, one competition I won't enter. Arguing with other kagglers isn't fun. This one will be ugly I'm afraid.\n\nI agree with @titericz , they should have imposed the use of GAN, or make it a swag only competition."
}
}![https://cdn.memegenerator.es/imagenes/memes/full/5/19/5191941.jpg](\"https://cdn.memegenerator.es/imagenes/memes/full/5/19/5191941.jpg\"){kind=link}