[Help] My generated images failed

[SaranchaiAng]

I just follow the generation guide with some modify(CUDA) as shown in the following code (I am quite new in deep learning)

set CUDA_VISIBLE_DEVICES=0 & python generate.py --ckpt_source ./checkpoints/ffhq_sketches.pt --ckpt_target ./checkpoints/ffhq_sketches.pt --load_noise noise.pt

And the attached image is the generated image I got.
I just wondering if I did something wrong? Any recommendation?

Thanks

[gouayao]

Your pre-training source domain model should be source_ffhq.pt.

[SaranchaiAng]

Your pre-training source domain model should be source_ffhq.pt.

Thank you for your answer @gouayao. I do fix that but I still got preliminary face outline not an actual face as the image below, it looks like an unfinished generation. I attached the generate.py arguments default settings below. Please help thanks

[gouayao]

Can I have a look at your code command?

[SaranchaiAng]

Can I have a look at your code command?
Thank you @gouayao

First, I try out this code
set CUDA_VISIBLE_DEVICES=0 & python generate.py --ckpt_target checkpoints/source_ffhq.pt
and this is the generated image I got

I tried another code
set CUDA_VISIBLE_DEVICES=0 & python generate.py --ckpt_target checkpoints/ffhq_sketches.pt --load_noise noise.pt
and this is the result I got

Thanks for any advice.

[gouayao]

I recommend that you print the pixel value of the output image to see if it is at (-1, 1). I replicated it on my computer and it was correct. Maybe you can experiment again with the source code.

[SaranchaiAng]

I recommend that you print the pixel value of the output image to see if it is at (-1, 1). I replicated it on my computer and it was correct. Maybe you can experiment again with the source code.

Hello brother thank you for all of your help.

I finally make it works.

However do you know if there is any way to snap the training model and continue the training later because my PC has insufficient GPU memory so I need to execute it on google colab and it keeps disconnecting after training.

Thanks

[gouayao]

You need about 24 GB GPU memory(3090, RTX). Or you can apply for a paid computer service. I'm sorry. I don't have a better way.

[SaranchaiAng]

You need about 24 GB GPU memory(3090, RTX). Or you can apply for a paid computer service. I'm sorry. I don't have a better way.

Yeah. I paid for the colab pro now. Everything works fine and it trains much faster than the free colab.
Lastly, thanks a lot again brother.

[gouayao]

You're welcome. Good Luck!

…

---- 回复的原邮件 ---- | 发件人 | Saranchai ***@***.***> | | 日期 | 2022年05月11日 19:49 | | 收件人 | ***@***.***> | | 抄送至 | ***@***.******@***.***> | | 主题 | Re: [utkarshojha/few-shot-gan-adaptation] [Help] My generated images failed (Issue #33) | You need about 24 GB GPU memory(3090, RTX). Or you can apply for a paid computer service. I'm sorry. I don't have a better way. Yeah. I paid for the colab pro now. Everything works fine and it trains much faster than the free colab. Lastly, thanks a lot again brother. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>