For example, the paper”Generating Diverse High-Fidelity Images with VQ-VAE-2” offers an alternative solution to gan. Are such methods allowed?
\n\nI can image one first learn the latent encoding space. Then the train images becomes the vector quantisation center. We can perturb these centers (prototype vectors) to generate new novel samples. The decoder generate realistic images from these encoded latent samples.
\n\nWhether the solution memorising the train images depends on how much we perturbed from the prototype vectors. But this is difficult to judge.
\n\nAlso, as a side note, if we use inception network as encoder, would it lead to better kaggle score?
", "messages": [ { "id": "595796", "postDate": "08/09/2019 17:26:32", "content": "For example, the paper”Generating Diverse High-Fidelity Images with VQ-VAE-2” offers an alternative solution to gan. Are such methods allowed?
\n\nI can image one first learn the latent encoding space. Then the train images becomes the vector quantisation center. We can perturb these centers (prototype vectors) to generate new novel samples. The decoder generate realistic images from these encoded latent samples.
\n\nWhether the solution memorising the train images depends on how much we perturbed from the prototype vectors. But this is difficult to judge.
\n\nAlso, as a side note, if we use inception network as encoder, would it lead to better kaggle score?
", "rawMarkdown": "For example, the paper”Generating Diverse High-Fidelity Images with VQ-VAE-2” offers an alternative solution to gan. Are such methods allowed?\n\nI can image one first learn the latent encoding space. Then the train images becomes the vector quantisation center. We can perturb these centers (prototype vectors) to generate new novel samples. The decoder generate realistic images from these encoded latent samples.\n\n\nWhether the solution memorising the train images depends on how much we perturbed from the prototype vectors. But this is difficult to judge.\n\nAlso, as a side note, if we use inception network as encoder, would it lead to better kaggle score?", "votes": null }, { "id": "595891", "postDate": "08/09/2019 19:52:32", "content": "As the competition description reads:\n\nImportant 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\nAnd here also:\n\nIn this competition, you’ll be training **generative models** to create images of dogs. Only this time… there’s no ground truth data for you to predict. \n\nI would suppose that it is not only about GANs.
Also, here is a discussion link that explicitly states:\n\nThis competition is indeed designed for GANs. However, Kaggle doesn't want to limit the solution algorithms one can use to solve a problem. For example, VAE methods such as this paper seem very promising, and we would love to see if they perform better than GANs! Therefore, we didn't want to specifically limit participants to use GANs in our official rules. However, we do want to discourage the use of \"altering training images\" - see #2.\n
@hengck23 - the rules don't allow inception or any other externel data, and no internet connection either - see Wendy's reply in Welcome(feedback) https://www.kaggle.com/c/generative-dog-images/discussion/97751#564046
\n\nCPU Kernel <= 9 hours run-time with 16 GB RAM globally available\n GPU Kernel <= 9 hours run-time with 13 GB RAM globally available (Turn on via kernel \"Settings\" menu)\n No TPU support in Kernels, currently\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 No custom packages enabled in kernels
\n\nI've been working on a pytorch version of VQ-VAE-2 and with the RAM, time and other restrictions like no datasets or custom packages it is hard going, but interesting anyways and good to learn more about other generative models. I think even if you could get a good model working, the amount of time it takes to generate the 10K samples for submission is quite long as well. Having said all that I'm sure you could come up with some great ideas here! Good luck.
", "rawMarkdown": "hengck23 - the rules don't allow inception or any other externel data, and no internet connection either - see Wendy's reply in Welcome(feedback) https://www.kaggle.com/c/generative-dog-images/discussion/97751#564046\n\n CPU Kernel <= 9 hours run-time with 16 GB RAM globally available\n GPU Kernel <= 9 hours run-time with 13 GB RAM globally available (Turn on via kernel \"Settings\" menu)\n No TPU support in Kernels, currently\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 No custom packages enabled in kernels\n \nI've been working on a pytorch version of VQ-VAE-2 and with the RAM, time and other restrictions like no datasets or custom packages it is hard going, but interesting anyways and good to learn more about other generative models. I think even if you could get a good model working, the amount of time it takes to generate the 10K samples for submission is quite long as well. Having said all that I'm sure you could come up with some great ideas here! Good luck.", "votes": null } ], "comments": [ { "id": 595891, "author_name": "purplejester", "author_url": "", "post_date": "08/09/2019 19:52:32", "content": "As the competition description reads:\n\nImportant 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\nAnd here also:\n\nIn this competition, you’ll be training **generative models** to create images of dogs. Only this time… there’s no ground truth data for you to predict. \n\nI would suppose that it is not only about GANs.
Also, here is a discussion link that explicitly states:\n\nThis competition is indeed designed for GANs. However, Kaggle doesn't want to limit the solution algorithms one can use to solve a problem. For example, VAE methods such as this paper seem very promising, and we would love to see if they perform better than GANs! Therefore, we didn't want to specifically limit participants to use GANs in our official rules. However, we do want to discourage the use of \"altering training images\" - see #2.\n
@hengck23 - the rules don't allow inception or any other externel data, and no internet connection either - see Wendy's reply in Welcome(feedback) https://www.kaggle.com/c/generative-dog-images/discussion/97751#564046
\n\nCPU Kernel <= 9 hours run-time with 16 GB RAM globally available\n GPU Kernel <= 9 hours run-time with 13 GB RAM globally available (Turn on via kernel \"Settings\" menu)\n No TPU support in Kernels, currently\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 No custom packages enabled in kernels
\n\nI've been working on a pytorch version of VQ-VAE-2 and with the RAM, time and other restrictions like no datasets or custom packages it is hard going, but interesting anyways and good to learn more about other generative models. I think even if you could get a good model working, the amount of time it takes to generate the 10K samples for submission is quite long as well. Having said all that I'm sure you could come up with some great ideas here! Good luck.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "595796": "For example, the paper”Generating Diverse High-Fidelity Images with VQ-VAE-2” offers an alternative solution to gan. Are such methods allowed?\n\nI can image one first learn the latent encoding space. Then the train images becomes the vector quantisation center. We can perturb these centers (prototype vectors) to generate new novel samples. The decoder generate realistic images from these encoded latent samples.\n\n\nWhether the solution memorising the train images depends on how much we perturbed from the prototype vectors. But this is difficult to judge.\n\nAlso, as a side note, if we use inception network as encoder, would it lead to better kaggle score?", "595891": "As the competition description reads:\n```\nImportant 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```\nAnd here also:\n```\nIn this competition, you’ll be training **generative models** to create images of dogs. Only this time… there’s no ground truth data for you to predict. \n```\nI would suppose that it is not only about GANs.\n\nAlso, here is [a discussion link](https://www.kaggle.com/c/generative-dog-images/discussion/98183#latest-595766) that explicitly states:\n```\nThis competition is indeed designed for GANs. However, Kaggle doesn't want to limit the solution algorithms one can use to solve a problem. For example, VAE methods such as this paper seem very promising, and we would love to see if they perform better than GANs! Therefore, we didn't want to specifically limit participants to use GANs in our official rules. However, we do want to discourage the use of \"altering training images\" - see #2.\n```", "596042": "hengck23 - the rules don't allow inception or any other externel data, and no internet connection either - see Wendy's reply in Welcome(feedback) https://www.kaggle.com/c/generative-dog-images/discussion/97751#564046\n\n CPU Kernel <= 9 hours run-time with 16 GB RAM globally available\n GPU Kernel <= 9 hours run-time with 13 GB RAM globally available (Turn on via kernel \"Settings\" menu)\n No TPU support in Kernels, currently\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 No custom packages enabled in kernels\n \nI've been working on a pytorch version of VQ-VAE-2 and with the RAM, time and other restrictions like no datasets or custom packages it is hard going, but interesting anyways and good to learn more about other generative models. I think even if you could get a good model working, the amount of time it takes to generate the 10K samples for submission is quite long as well. Having said all that I'm sure you could come up with some great ideas here! Good luck." }, "source": "meta" }