I want to train the model for African Ethnicity on Colab TPU. But when I do it, "from op import FusedLeakyReLU, fused_leaky_relu, upfirdn2d" is wrong (No CUDA Home ./local/..) in "model.py".

I guess it may be an issue in folder "op" since it defines c++ extension.

would you mind taking some advice for me ?

I'm training a model for African projector model

I started the training, assuming --checkpoint_path_E is encoder.pt --checkpoint_path_D is stylegan2-ffhq-config-f.pt --checkpoint_path_P is 79999_iter.pth

I set --save_interval to 1000, so every 1000 steps the model will get saved.

Now, In the --exp_dir we will have the saved model in checkpoints folder and I think that model we can use as projector_African.pt

I tried inference by setting --checkpoint_path_E is encoder.pt --checkpoint_path_D= projector_African.pt --checkpoint_path_P is 79999_iter.pth

I got below error: KeyError 'g_ema' not found

I checked the projector_African.pt keys and I got these : dict_keys(['state_dict', 'opts'])

while I checked your provided projector_EastAsian.pt keys I got these: dict_keys(['g_ema', 'latent_avg'])

Why both models have different keys? Am I missing something while training?

Hello, Thanks for this amazing repo.

I'm able to do inference using available pretrained models but not able to start training due to below issue:

I think 3 checkpoints are required which are mentioned in the below file: https://github.com/a312863063/Model-Swap-Face_v2/blob/main/models/psp.py#L25

I couldn't find them in the below argument files: https://github.com/a312863063/Model-Swap-Face_v2/blob/main/options/train_options.py

What are these 3 checkpoints while training? encoder_dict = torch.load(args.checkpoint_path_E, map_location='cpu') # {'state_dict', 'opts'} decoder_dict = torch.load(args.checkpoint_path_D, map_location='cpu') # {'g_ema', 'latent_avg'} parser_dict = torch.load(args.checkpoint_path_P, map_location='cpu')