{ "id": 178899, "title": "Fast.ai Lecture 12: Generative Adverserial Networks (GANs)", "url": "/competitions/gan-getting-started/discussion/178899", "author_name": "James McGuigan", "post_date": "2020-08-31T19:40:32.039000", "votes": 9, "comment_count": 5, "views": 0, "content": "

This 2-hour lecture by Jeremy Howard (Kaggle Grandmaster), covers the theory and practice behind GANs and also covers the concept of Cycle Loss in a CycleGAN

\n
\n

We start today with a deep dive into the DarkNet architecture used in YOLOv3, and use it to better understand all the details and choices that you can make when implementing a resnet-ish architecture. The basic approach discussed here is what we used to win the DAWNBench competition!

\n

Then we’ll learn about Generative Adversarial Networks (GANs). This is, at its heart, a different kind of loss function. GANs have a generator and a discriminator that battle it out, and in the process combine to create a generative model that can create highly realistic outputs. We’ll be looking at the Wasserstein GAN variant, since it’s easier to train and more resilient to a range of hyperparameters.

\n
\n", "messages": [ { "id": 993268, "postDate": "2020-08-31T19:40:32.040Z", "content": "

This 2-hour lecture by Jeremy Howard (Kaggle Grandmaster), covers the theory and practice behind GANs and also covers the concept of Cycle Loss in a CycleGAN

\n
\n

We start today with a deep dive into the DarkNet architecture used in YOLOv3, and use it to better understand all the details and choices that you can make when implementing a resnet-ish architecture. The basic approach discussed here is what we used to win the DAWNBench competition!

\n

Then we’ll learn about Generative Adversarial Networks (GANs). This is, at its heart, a different kind of loss function. GANs have a generator and a discriminator that battle it out, and in the process combine to create a generative model that can create highly realistic outputs. We’ll be looking at the Wasserstein GAN variant, since it’s easier to train and more resilient to a range of hyperparameters.

\n
\n", "rawMarkdown": "This 2-hour lecture by Jeremy Howard (Kaggle Grandmaster), covers the theory and practice behind GANs and also covers the concept of Cycle Loss in a CycleGAN \n\n> We start today with a deep dive into the DarkNet architecture used in YOLOv3, and use it to better understand all the details and choices that you can make when implementing a resnet-ish architecture. The basic approach discussed here is what we used to win the DAWNBench competition!\n\n> Then we’ll learn about Generative Adversarial Networks (GANs). This is, at its heart, a different kind of loss function. GANs have a generator and a discriminator that battle it out, and in the process combine to create a generative model that can create highly realistic outputs. We’ll be looking at the Wasserstein GAN variant, since it’s easier to train and more resilient to a range of hyperparameters.\n\n- Video: https://course18.fast.ai/lessons/lesson12.html\n- Wiki: https://forums.fast.ai/t/part-2-lesson-12-wiki/15023", "votes": 9 }, { "id": 993272, "postDate": "2020-08-31T19:44:34.720Z", "content": "

This is great, thanks!

\n

There is no equivalent video using fastai v2 (at lest not yet). Right?

", "rawMarkdown": "This is great, thanks!\n\nThere is no equivalent video using fastai v2 (at lest not yet). Right?", "replies": [ { "id": 993274, "postDate": "2020-08-31T19:47:14.060Z", "content": "

There is a GAN module within the Fastai library:

\n", "rawMarkdown": "There is a GAN module within the Fastai library:\n- https://docs.fast.ai/vision.gan.html" }, { "id": 993293, "postDate": "2020-08-31T20:16:18.607Z", "content": "

Oh great, thanks!

", "rawMarkdown": "Oh great, thanks!" }, { "id": 993331, "postDate": "2020-08-31T21:01:41.930Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 994198, "postDate": "2020-09-01T13:02:23.370Z", "content": "

This is awesome Thank you!

", "rawMarkdown": "This is awesome Thank you!" } ], "comments": [ { "id": 993272, "author_name": "Santiago Viquez", "author_url": "", "post_date": "2020-08-31T19:44:34.720000", "content": "

This is great, thanks!

\n

There is no equivalent video using fastai v2 (at lest not yet). Right?

", "votes": 0, "replies": [ { "id": 993274, "author_name": "James McGuigan", "author_url": "", "post_date": "2020-08-31T19:47:14.060000", "content": "

There is a GAN module within the Fastai library:

\n", "votes": 0, "replies": [] }, { "id": 993293, "author_name": "Santiago Viquez", "author_url": "", "post_date": "2020-08-31T20:16:18.607000", "content": "

Oh great, thanks!

", "votes": 0, "replies": [] }, { "id": 993331, "author_name": "", "author_url": "", "post_date": "2020-08-31T21:01:41.930000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 994198, "author_name": "sai suraj", "author_url": "", "post_date": "2020-09-01T13:02:23.370000", "content": "

This is awesome Thank you!

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "993268": "This 2-hour lecture by Jeremy Howard (Kaggle Grandmaster), covers the theory and practice behind GANs and also covers the concept of Cycle Loss in a CycleGAN \n\n> We start today with a deep dive into the DarkNet architecture used in YOLOv3, and use it to better understand all the details and choices that you can make when implementing a resnet-ish architecture. The basic approach discussed here is what we used to win the DAWNBench competition!\n\n> Then we’ll learn about Generative Adversarial Networks (GANs). This is, at its heart, a different kind of loss function. GANs have a generator and a discriminator that battle it out, and in the process combine to create a generative model that can create highly realistic outputs. We’ll be looking at the Wasserstein GAN variant, since it’s easier to train and more resilient to a range of hyperparameters.\n\n- Video: https://course18.fast.ai/lessons/lesson12.html\n- Wiki: https://forums.fast.ai/t/part-2-lesson-12-wiki/15023", "993272": "This is great, thanks!\n\nThere is no equivalent video using fastai v2 (at lest not yet). Right?", "994198": "This is awesome Thank you!" } }