search

yuval-alaluf / SAM

Official Implementation for "Only a Matter of Style: Age Transformation Using a Style-Based Regression Model" (SIGGRAPH 2021) https://arxiv.org/abs/2102.02754
https://yuval-alaluf.github.io/SAM/
MIT License
632 stars 151 forks source link

Purpose of parameters for inference.py script #15

Closed MetiCodePrivateLimited closed 3 years ago

MetiCodePrivateLimited commented 3 years ago

Could you please guide us what is the purpose of following two parameters? I check details of Readme context but could not find relevant.

--test_batch_size=4 \ --test_workers=4 \

Also, how can we improve quality of old face effects?

yuval-alaluf commented 3 years ago

Please see my answer here: https://github.com/yuval-alaluf/SAM/issues/7#issuecomment-853137473

MetiCodePrivateLimited commented 3 years ago

how can we improve quality of old face effects?

yuval-alaluf commented 3 years ago

Can you be more specific? What are you looking to improve exactly?

MetiCodePrivateLimited commented 3 years ago

We want to apply effects of old face similar to faceapp android app.

MetiCodePrivateLimited commented 3 years ago

If you go for self trained datasets then from when we can arrange faces of people? Is there any website to provide such dataset?

vinduon commented 3 years ago

If you go for self trained datasets then from when we can arrange faces of people? Is there any website to provide such dataset?

You can go with FFHQ dataset or CelabA-HQ

yuval-alaluf commented 3 years ago

We want to apply effects of old face similar to faceapp android app.

Trying to capture aging affects such as gray hair is challenging. You can try lowering the L2 and LPIPS weights on the forward pass. This way the relative weight of the aging loss will increase and you are more likely to capture these changes. However, by lowering the L2 and LPIPS losses you may hurt the reconstruction quality and identity preservation of the result. In any case, what you want requires re-training SAM (possibly starting from the checkpoint I provided) and playing around with the loss weights

MetiCodePrivateLimited commented 3 years ago

Is there any website from where I could buy strong datasets for old age effects.?

yuval-alaluf commented 3 years ago

Is there any website from where I could buy strong datasets for old age effects.?

Collecting sufficient paired data across a wide range of ages will probably be very difficult, if not impossible. I do not know of any websites offering such datasets.