search

openai / vdvae

Repository for the paper "Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images"
MIT License
436 stars 85 forks source link

error when load model #3

Open betterze opened 3 years ago

betterze commented 3 years ago

Dear OpenAI team,

Thank you for sharing with us this great implementation.

When I try to load FFHQ 1204 using command:

wget https://openaipublic.blob.core.windows.net/very-deep-vaes-assets/vdvae-assets/ffhq1024-iter-1700000-log.jsonl
wget https://openaipublic.blob.core.windows.net/very-deep-vaes-assets/vdvae-assets/ffhq1024-iter-1700000-model.th
wget https://openaipublic.blob.core.windows.net/very-deep-vaes-assets/vdvae-assets/ffhq1024-iter-1700000-model-ema.th
wget https://openaipublic.blob.core.windows.net/very-deep-vaes-assets/vdvae-assets/ffhq1024-iter-1700000-opt.th
python train.py --hps ffhq1024 --restore_path ffhq1024-iter-1700000-model.th --restore_ema_path ffhq1024-iter-1700000-model-ema.th --restore_log_path ffhq1024-iter-1700000-log.jsonl --restore_optimizer_path ffhq1024-iter-1700000-opt.th --test_eval

It output:

Missing key(s) in state_dict: "encoder.enc_blocks.68.c1.weight", "encoder.enc_blocks.68.c1.bias", "encoder.enc_blocks.68.c2.weight", "encoder.enc_blocks.68.c2.bias", "encoder.enc_blocks.68.c3.weight", "encoder.enc_blocks.68.c3.bias", "encoder.enc_blocks.68.c4.weight", "encoder.enc_blocks.68.c4.bias", "encoder.enc_blocks.69.c1.weight", "encoder.enc_blocks.69.c1.bias", "encoder.enc_blocks.69.c2.weight", "encoder.enc_blocks.69.c2.bias", "encoder.enc_blocks.69.c3.weight", "encoder.enc_blocks.69.c3.bias", "encoder.enc_blocks.69.c4.weight", "encoder.enc_blocks.69.c4.bias", "decoder.dec_blocks.66.enc.c1.weight", "decoder.dec_blocks.66.enc.c1.bias", "decoder.dec_blocks.66.enc.c2.weight", "decoder.dec_blocks.66.enc.c2.bias", "decoder.dec_blocks.66.enc.c3.weight", "decoder.dec_blocks.66.enc.c3.bias", "decoder.dec_blocks.66.enc.c4.weight", "decoder.dec_blocks.66.enc.c4.bias", "decoder.dec_blocks.66.prior.c1.weight", "decoder.dec_blocks.66.prior.c1.bias", "decoder.dec_blocks.66.prior.c2.weight", "decoder.dec_blocks.66.prior.c2.bias", "decoder.dec_blocks.66.prior.c3.weight", "decoder.dec_blocks.66.prior.c3.bias", "decoder.dec_blocks.66.prior.c4.weight", "decoder.dec_blocks.66.prior.c4.bias", "decoder.dec_blocks.66.z_proj.weight", "decoder.dec_blocks.66.z_proj.bias", "decoder.dec_blocks.66.resnet.c1.weight", "decoder.dec_blocks.66.resnet.c1.bias", "decoder.dec_blocks.66.resnet.c2.weight", "decoder.dec_blocks.66.resnet.c2.bias", "decoder.dec_blocks.66.resnet.c3.weight", "decoder.dec_blocks.66.resnet.c3.bias", "decoder.dec_blocks.66.resnet.c4.weight", "decoder.dec_blocks.66.resnet.c4.bias", "decoder.dec_blocks.67.enc.c1.weight", "decoder.dec_blocks.67.enc.c1.bias", "decoder.dec_blocks.67.enc.c2.weight", "decoder.dec_blocks.67.enc.c2.bias", "decoder.dec_blocks.67.enc.c3.weight", "decoder.dec_blocks.67.enc.c3.bias", "decoder.dec_blocks.67.enc.c4.weight", "decoder.dec_blocks.67.enc.c4.bias", "decoder.dec_blocks.67.prior.c1.weight", "decoder.dec_blocks.67.prior.c1.bias", "decoder.dec_blocks.67.prior.c2.weight", "decoder.dec_blocks.67.prior.c2.bias", "decoder.dec_blocks.67.prior.c3.weight", "decoder.dec_blocks.67.prior.c3.bias", "decoder.dec_blocks.67.prior.c4.weight", "decoder.dec_blocks.67.prior.c4.bias", "decoder.dec_blocks.67.z_proj.weight", "decoder.dec_blocks.67.z_proj.bias", "decoder.dec_blocks.67.resnet.c1.weight", "decoder.dec_blocks.67.resnet.c1.bias", "decoder.dec_blocks.67.resnet.c2.weight", "decoder.dec_blocks.67.resnet.c2.bias", "decoder.dec_blocks.67.resnet.c3.weight", "decoder.dec_blocks.67.resnet.c3.bias", "decoder.dec_blocks.67.resnet.c4.weight", "decoder.dec_blocks.67.resnet.c4.bias", "decoder.dec_blocks.68.enc.c1.weight", "decoder.dec_blocks.68.enc.c1.bias", "decoder.dec_blocks.68.enc.c2.weight", "decoder.dec_blocks.68.enc.c2.bias", "decoder.dec_blocks.68.enc.c3.weight", "decoder.dec_blocks.68.enc.c3.bias", "decoder.dec_blocks.68.enc.c4.weight", "decoder.dec_blocks.68.enc.c4.bias", "decoder.dec_blocks.68.prior.c1.weight", "decoder.dec_blocks.68.prior.c1.bias", "decoder.dec_blocks.68.prior.c2.weight", "decoder.dec_blocks.68.prior.c2.bias", "decoder.dec_blocks.68.prior.c3.weight", "decoder.dec_blocks.68.prior.c3.bias", "decoder.dec_blocks.68.prior.c4.weight", "decoder.dec_blocks.68.prior.c4.bias", "decoder.dec_blocks.68.z_proj.weight", "decoder.dec_blocks.68.z_proj.bias", "decoder.dec_blocks.68.resnet.c1.weight", "decoder.dec_blocks.68.resnet.c1.bias", "decoder.dec_blocks.68.resnet.c2.weight", "decoder.dec_blocks.68.resnet.c2.bias", "decoder.dec_blocks.68.resnet.c3.weight", "decoder.dec_blocks.68.resnet.c3.bias", "decoder.dec_blocks.68.resnet.c4.weight", "decoder.dec_blocks.68.resnet.c4.bias", "decoder.dec_blocks.69.enc.c1.weight", "decoder.dec_blocks.69.enc.c1.bias", "decoder.dec_blocks.69.enc.c2.weight", "decoder.dec_blocks.69.enc.c2.bias", "decoder.dec_blocks.69.enc.c3.weight", "decoder.dec_blocks.69.enc.c3.bias", "decoder.dec_blocks.69.enc.c4.weight", "decoder.dec_blocks.69.enc.c4.bias", "decoder.dec_blocks.69.prior.c1.weight", "decoder.dec_blocks.69.prior.c1.bias", "decoder.dec_blocks.69.prior.c2.weight", "decoder.dec_blocks.69.prior.c2.bias", "decoder.dec_blocks.69.prior.c3.weight", "decoder.dec_blocks.69.prior.c3.bias", "decoder.dec_blocks.69.prior.c4.weight", "decoder.dec_blocks.69.prior.c4.bias", "decoder.dec_blocks.69.z_proj.weight", "decoder.dec_blocks.69.z_proj.bias", "decoder.dec_blocks.69.resnet.c1.weight", "decoder.dec_blocks.69.resnet.c1.bias", "decoder.dec_blocks.69.resnet.c2.weight", "decoder.dec_blocks.69.resnet.c2.bias", "decoder.dec_blocks.69.resnet.c3.weight", "decoder.dec_blocks.69.resnet.c3.bias", "decoder.dec_blocks.69.resnet.c4.weight", "decoder.dec_blocks.69.resnet.c4.bias". 
    size mismatch for encoder.in_conv.weight: copying a param with shape torch.Size([512, 3, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 3, 3, 3]).
    size mismatch for encoder.in_conv.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.0.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([4, 16, 1, 1]).
    size mismatch for encoder.enc_blocks.0.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([4]).
    size mismatch for encoder.enc_blocks.0.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([4, 4, 3, 3]).
    size mismatch for encoder.enc_blocks.0.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([4]).
    size mismatch for encoder.enc_blocks.0.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([4, 4, 3, 3]).
    size mismatch for encoder.enc_blocks.0.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([4]).
    size mismatch for encoder.enc_blocks.0.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 4, 1, 1]).
    size mismatch for encoder.enc_blocks.0.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.1.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([4, 16, 1, 1]).
    size mismatch for encoder.enc_blocks.1.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([4]).
    size mismatch for encoder.enc_blocks.1.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([4, 4, 3, 3]).
    size mismatch for encoder.enc_blocks.1.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([4]).
    size mismatch for encoder.enc_blocks.1.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([4, 4, 3, 3]).
    size mismatch for encoder.enc_blocks.1.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([4]).
    size mismatch for encoder.enc_blocks.1.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 4, 1, 1]).
    size mismatch for encoder.enc_blocks.1.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.2.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([8, 32, 1, 1]).
    size mismatch for encoder.enc_blocks.2.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.2.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 8, 3, 3]).
    size mismatch for encoder.enc_blocks.2.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.2.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 8, 3, 3]).
    size mismatch for encoder.enc_blocks.2.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.2.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 8, 1, 1]).
    size mismatch for encoder.enc_blocks.2.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([32]).
    size mismatch for encoder.enc_blocks.3.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([8, 32, 1, 1]).
    size mismatch for encoder.enc_blocks.3.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.3.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 8, 3, 3]).
    size mismatch for encoder.enc_blocks.3.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.3.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 8, 3, 3]).
    size mismatch for encoder.enc_blocks.3.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.3.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 8, 1, 1]).
    size mismatch for encoder.enc_blocks.3.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([32]).
    size mismatch for encoder.enc_blocks.4.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([8, 32, 1, 1]).
    size mismatch for encoder.enc_blocks.4.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.4.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 8, 3, 3]).
    size mismatch for encoder.enc_blocks.4.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.4.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 8, 3, 3]).
    size mismatch for encoder.enc_blocks.4.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.4.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 8, 1, 1]).
    size mismatch for encoder.enc_blocks.4.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([32]).
    size mismatch for encoder.enc_blocks.5.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([8, 32, 1, 1]).
    size mismatch for encoder.enc_blocks.5.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.5.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 8, 3, 3]).
    size mismatch for encoder.enc_blocks.5.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.5.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 8, 3, 3]).
    size mismatch for encoder.enc_blocks.5.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([8]).
    size mismatch for encoder.enc_blocks.5.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 8, 1, 1]).
    size mismatch for encoder.enc_blocks.5.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([32]).
    size mismatch for encoder.enc_blocks.6.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 64, 1, 1]).
    size mismatch for encoder.enc_blocks.6.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.6.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.6.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.6.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.6.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.6.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 16, 1, 1]).
    size mismatch for encoder.enc_blocks.6.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([64]).
    size mismatch for encoder.enc_blocks.7.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 64, 1, 1]).
    size mismatch for encoder.enc_blocks.7.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.7.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.7.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.7.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.7.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.7.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 16, 1, 1]).
    size mismatch for encoder.enc_blocks.7.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([64]).
    size mismatch for encoder.enc_blocks.8.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 64, 1, 1]).
    size mismatch for encoder.enc_blocks.8.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.8.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.8.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.8.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.8.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.8.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 16, 1, 1]).
    size mismatch for encoder.enc_blocks.8.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([64]).
    size mismatch for encoder.enc_blocks.9.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 64, 1, 1]).
    size mismatch for encoder.enc_blocks.9.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.9.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.9.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.9.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.9.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.9.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 16, 1, 1]).
    size mismatch for encoder.enc_blocks.9.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([64]).
    size mismatch for encoder.enc_blocks.10.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 64, 1, 1]).
    size mismatch for encoder.enc_blocks.10.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.10.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.10.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.10.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.10.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.10.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 16, 1, 1]).
    size mismatch for encoder.enc_blocks.10.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([64]).
    size mismatch for encoder.enc_blocks.11.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 64, 1, 1]).
    size mismatch for encoder.enc_blocks.11.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.11.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.11.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.11.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for encoder.enc_blocks.11.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for encoder.enc_blocks.11.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 16, 1, 1]).
    size mismatch for encoder.enc_blocks.11.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([64]).
    size mismatch for encoder.enc_blocks.64.c2.weight: copying a param with shape torch.Size([128, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 128, 3, 3]).
    size mismatch for encoder.enc_blocks.64.c3.weight: copying a param with shape torch.Size([128, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 128, 3, 3]).
    size mismatch for encoder.enc_blocks.65.c2.weight: copying a param with shape torch.Size([128, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 128, 3, 3]).
    size mismatch for encoder.enc_blocks.65.c3.weight: copying a param with shape torch.Size([128, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 128, 3, 3]).
    size mismatch for decoder.gain: copying a param with shape torch.Size([1, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 16, 1, 1]).
    size mismatch for decoder.bias: copying a param with shape torch.Size([1, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 16, 1, 1]).
    size mismatch for decoder.dec_blocks.65.enc.c1.weight: copying a param with shape torch.Size([128, 1024, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 128, 1, 1]).
    size mismatch for decoder.dec_blocks.65.enc.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.enc.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for decoder.dec_blocks.65.enc.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.enc.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for decoder.dec_blocks.65.enc.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.enc.c4.weight: copying a param with shape torch.Size([32, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 16, 1, 1]).
    size mismatch for decoder.dec_blocks.65.prior.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 64, 1, 1]).
    size mismatch for decoder.dec_blocks.65.prior.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.prior.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for decoder.dec_blocks.65.prior.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.prior.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for decoder.dec_blocks.65.prior.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.prior.c4.weight: copying a param with shape torch.Size([544, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([96, 16, 1, 1]).
    size mismatch for decoder.dec_blocks.65.prior.c4.bias: copying a param with shape torch.Size([544]) from checkpoint, the shape in current model is torch.Size([96]).
    size mismatch for decoder.dec_blocks.65.z_proj.weight: copying a param with shape torch.Size([512, 16, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 16, 1, 1]).
    size mismatch for decoder.dec_blocks.65.z_proj.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([64]).
    size mismatch for decoder.dec_blocks.65.resnet.c1.weight: copying a param with shape torch.Size([128, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([16, 64, 1, 1]).
    size mismatch for decoder.dec_blocks.65.resnet.c1.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.resnet.c2.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for decoder.dec_blocks.65.resnet.c2.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.resnet.c3.weight: copying a param with shape torch.Size([128, 128, 3, 3]) from checkpoint, the shape in current model is torch.Size([16, 16, 3, 3]).
    size mismatch for decoder.dec_blocks.65.resnet.c3.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([16]).
    size mismatch for decoder.dec_blocks.65.resnet.c4.weight: copying a param with shape torch.Size([512, 128, 1, 1]) from checkpoint, the shape in current model is torch.Size([64, 16, 1, 1]).
    size mismatch for decoder.dec_blocks.65.resnet.c4.bias: copying a param with shape torch.Size([512]) from checkpoint, the shape in current model is torch.Size([64]).
    size mismatch for decoder.out_net.out_conv.weight: copying a param with shape torch.Size([100, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([20, 16, 1, 1]).
    size mismatch for decoder.out_net.out_conv.bias: copying a param with shape torch.Size([100]) from checkpoint, the shape in current model is torch.Size([20]).

It seems like the model and weights are not compatible. Could you tell me how to solve this problem?

Thank you very much for your help.

Best Wishes,

Alex

rewonc commented 3 years ago

Hi Alex,

Investigating this -- I think I somehow overwrote the weights for the 1024x1024 model with the 256x256 model (probably because they were named pretty similarly). That's why the sizes are mismatched -- it actually loads the ffhq256 model. I'll see if I can get the correct version -- but it might take some time, sorry about that.

Rewon

betterze commented 3 years ago

Thx for clarification.

IlianHara commented 3 years ago

Hi Rewon,

Just to follow-up, were you able to upload the correct version of the 1024x1024 model?

Thanks for sharing, Ilian