search

lucidrains / lightweight-gan

Implementation of 'lightweight' GAN, proposed in ICLR 2021, in Pytorch. High resolution image generations that can be trained within a day or two
MIT License
1.62k stars 220 forks source link

Error(s) in loading state_dict for LightweightGAN #116

Open rickyars opened 2 years ago

rickyars commented 2 years ago

When using --antialias I get the following error when trying to call --show-progress:

Traceback (most recent call last): File "/usr/local/bin/lightweight_gan", line 8, in sys.exit(main()) File "/usr/local/lib/python3.7/dist-packages/lightweight_gan/cli.py", line 190, in main fire.Fire(train_from_folder) File "/usr/local/lib/python3.7/dist-packages/fire/core.py", line 141, in Fire component_trace = _Fire(component, args, parsed_flag_args, context, name) File "/usr/local/lib/python3.7/dist-packages/fire/core.py", line 471, in _Fire target=component.name) File "/usr/local/lib/python3.7/dist-packages/fire/core.py", line 681, in _CallAndUpdateTrace component = fn(*varargs, *kwargs) File "/usr/local/lib/python3.7/dist-packages/lightweight_gan/cli.py", line 171, in train_from_folder model.show_progress(num_images=num_image_tiles, types=generate_types) File "/usr/local/lib/python3.7/dist-packages/torch/autograd/grad_mode.py", line 28, in decorate_context return func(args, **kwargs) File "/usr/local/lib/python3.7/dist-packages/lightweight_gan/lightweight_gan.py", line 1306, in show_progress self.load(checkpoint, print_version=False) File "/usr/local/lib/python3.7/dist-packages/lightweight_gan/lightweight_gan.py", line 1471, in load raise e File "/usr/local/lib/python3.7/dist-packages/lightweight_gan/lightweight_gan.py", line 1468, in load self.GAN.load_state_dict(load_data['GAN']) File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1483, in load_state_dict self.class.name, "\n\t".join(error_msgs))) RuntimeError: Error(s) in loading state_dict for LightweightGAN: Unexpected key(s) in state_dict: "G.layers.0.0.1.f", "G.layers.1.0.1.f", "G.layers.2.0.1.f", "G.layers.3.0.1.f", "G.layers.4.0.1.f", "G.layers.5.0.1.f", "G.layers.6.0.1.f", "D.non_residual_layers.0.0.f", "D.residual_layers.0.0.branches.0.0.f", "D.residual_layers.0.0.branches.1.0.f", "D.residual_layers.1.0.branches.0.0.f", "D.residual_layers.1.0.branches.1.0.f", "D.residual_layers.2.0.branches.0.0.f", "D.residual_layers.2.0.branches.1.0.f", "D.residual_layers.3.0.branches.0.0.f", "D.residual_layers.3.0.branches.1.0.f", "D.residual_layers.4.0.branches.0.0.f", "D.residual_layers.4.0.branches.1.0.f", "D.to_logits.0.f", "D.to_shape_disc_out.2.branches.0.0.f", "D.to_shape_disc_out.2.branches.1.0.f", "GE.layers.0.0.1.f", "GE.layers.1.0.1.f", "GE.layers.2.0.1.f", "GE.layers.3.0.1.f", "GE.layers.4.0.1.f", "GE.layers.5.0.1.f", "GE.layers.6.0.1.f", "D_aug.D.non_residual_layers.0.0.f", "D_aug.D.residual_layers.0.0.branches.0.0.f", "D_aug.D.residual_layers.0.0.branches.1.0.f", "D_aug.D.residual_layers.1.0.branches.0.0.f", "D_aug.D.residual_layers.1.0.branches.1.0.f", "D_aug.D.residual_layers.2.0.branches.0.0.f", "D_aug.D.residual_layers.2.0.branches.1.0.f", "D_aug.D.residual_layers.3.0.branches.0.0.f", "D_aug.D.residual_layers.3.0.branches.1.0.f", "D_aug.D.residual_layers.4.0.branches.0.0.f", "D_aug.D.residual_layers.4.0.branches.1.0.f", "D_aug.D.to_logits.0.f", "D_aug.D.to_shape_disc_out.2.branches.0.0.f", "D_aug.D.to_shape_disc_out.2.branches.1.0.f"

iScriptLex commented 2 years ago

You can't change antialiasing type when network is already exists and has checkpoints. You should start entirely new instance of network for switching this feature, because it inflicts the network structure.

rickyars commented 2 years ago

Wow - I'm such an idiot. Thank you.