search

Octavian-ai / BigGAN-TPU-TensorFlow

TPU enabled Tensorflow implementation of "Large Scale GAN Training for High Fidelity Natural Image Synthesis" (BigGAN)
https://octavian.ai
MIT License
34 stars 13 forks source link

Project in working condition? #1

Closed apoorvkh closed 5 years ago

apoorvkh commented 5 years ago

Hi,

I'm trying to train BigGAN on a dataset and see this is a recent project -- I have a few questions. Have you had success/progress with training on ImageNet so far? Are there any specific training scripts that are or are not currently working? What tpu-size are you using?

Thanks!

davidhughhenrymack commented 5 years ago

Hi, Thanks for the questions!

I’ve been debugging the model, and am now running it on ImageNet again to see how well it does.

The best way for you to demo it is to run the MNIST or cifar launch scripts. They’ll quickly train to reasonable looking results (e.g. MNIST perfectly conditions and makes nice looking digits).

I’m generally running on a single TPU (=8 cores) as that’s all I can get my hands on

Does that help? Can I answer any more / encourage you to contribute?

On March 16, 2019 at 3:10:16 PM, Apoorv Khandelwal (notifications@github.com) wrote:

Hi,

I'm trying to train BigGAN on a dataset and see this is a recent project -- I have a few questions. Have you had success/progress with training on ImageNet so far? Are there any specific training scripts that are or are not currently working? What tpu-size are you using?

Thanks!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Octavian-ai/BigGAN-TPU-TensorFlow/issues/1, or mute the thread https://github.com/notifications/unsubscribe-auth/AOUFIuKsvF145fs2UpovV8tINDMvJMe3ks5vXWvIgaJpZM4b4DnL .

davidhughhenrymack commented 5 years ago

Oh one more thing - you can see my training runs (live and previous) here: https://www.comet.ml/davidhughhenrymack/biggan

On March 16, 2019 at 3:15:34 PM, David Mack (davidhughhenrymack@gmail.com) wrote:

Hi, Thanks for the questions!

I’ve been debugging the model, and am now running it on ImageNet again to see how well it does.

The best way for you to demo it is to run the MNIST or cifar launch scripts. They’ll quickly train to reasonable looking results (e.g. MNIST perfectly conditions and makes nice looking digits).

I’m generally running on a single TPU (=8 cores) as that’s all I can get my hands on

Does that help? Can I answer any more / encourage you to contribute?

On March 16, 2019 at 3:10:16 PM, Apoorv Khandelwal (notifications@github.com) wrote:

Hi,

I'm trying to train BigGAN on a dataset and see this is a recent project -- I have a few questions. Have you had success/progress with training on ImageNet so far? Are there any specific training scripts that are or are not currently working? What tpu-size are you using?

Thanks!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Octavian-ai/BigGAN-TPU-TensorFlow/issues/1, or mute the thread https://github.com/notifications/unsubscribe-auth/AOUFIuKsvF145fs2UpovV8tINDMvJMe3ks5vXWvIgaJpZM4b4DnL .

apoorvkh commented 5 years ago

Thanks! I'll let you know if I decide to try it out. One more question - is that a v2 or v3 TPU?

davidhughhenrymack commented 5 years ago

I used the v2.8 as it’s all I have access to. I want to get access to a pod

On March 16, 2019 at 4:13:21 PM, Apoorv Khandelwal (notifications@github.com) wrote:

Thanks! I'll let you know if I decide to try it out. One more question - is that a v2 or v3 TPU?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Octavian-ai/BigGAN-TPU-TensorFlow/issues/1#issuecomment-473600390, or mute the thread https://github.com/notifications/unsubscribe-auth/AOUFIh3qwkZmu-Q1gVIcDMt7k3kbgq9-ks5vXXqQgaJpZM4b4DnL .

apoorvkh commented 5 years ago

Sounds good. Thanks!