The swap result of this model is not good.
The similarity of the source face with the swapped face is very poor.

Note that-> I have used proper alignment methods as mentioned in this repository.

I will upload the similarity matrix and result later.

Hi @mapooon , thanks for your work! May I know how to align the images? Whether it same with the arcface do as usual?

self.Decoder_inchannel = [1024, 2048, 1024, 512, 256, 128]
self.Decoder_outchannel = [1024, 512, 256, 128, 64, 32]
self.Decoder = nn.ModuleDict({f'layer_{i}' : nn.Sequential(
nn.ConvTranspose2d(self.Decoder_inchannel[i], self.Decoder_outchannel[i], kernel_size=4, stride=2, padding=1),
nn.BatchNorm2d(self.Decoder_outchannel[i]),
nn.LeakyReLU(0.1)
)for i in range(6)})

self.Decoder_inchannel1 is not equal to self.Decoder_outchannel0.is there a problem here
i think there is an issue with setting the number of channels.

@mapooon
Thank you for your work.
I have been testing out your face swap model and the results seems to be quite bad.

steps taken:

1. download blendswap.pth
2. use insightface detector and face_align in their utils on both source and target
3. change target to (256, 256) when converting to tensor
4. result (supposed to be emma on cage lol)
   

Any advice on how to make it work properly?

Hello author, thank you for your sharing, can the proposed encoder be applied to StyleGAN?

I want to know the code for computing 'Identity Distance' and 'Attribute Distance.'

Thank you.

Is there a specific size requirement for the source and target images? I successfully swapped faces using the provided source (112x112) and target (256x256). However, when I resized the source image from 112x112 to 256x256, an error occurred.

I have no name!@cbf99791397b:/workspace/swapping$ python3 inference.py -w /workspace/checkpoints/blendswap.pth -s examples/source.png -t examples/target.png -o examples/output.png
Traceback (most recent call last):
  File "/workspace/swapping/inference.py", line 31, in <module>
    output = model(target_img, source_img)
  File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
    return forward_call(*input, **kwargs)
  File "/workspace/swapping/blendswap.py", line 60, in forward
    z_id = self.Z_e(source_img)
  File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
    return forward_call(*input, **kwargs)
  File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/container.py", line 204, in forward
    input = module(input)
  File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
    return forward_call(*input, **kwargs)
  File "/workspace/swapping/model/iresnet.py", line 151, in forward
    x = self.fc(x.float() if self.fp16 else x)
  File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
    return forward_call(*input, **kwargs)
  File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/linear.py", line 114, in forward
    return F.linear(input, self.weight, self.bias)
RuntimeError: mat1 and mat2 shapes cannot be multiplied (1x131072 and 25088x512)

great work but where face swap code

I notice that the result depends on how well the face is aligned, there is some script to align faces, it would be nice to be able to swap faces when the faces are not aligned

Hi, great work! I wanna know whether you will release the whole model(BlendFace + face-swapping model).

Figure 9 in your paper visualizes the images of ArcFace and BleedFace. If convenient, could you please provide some details? I really appreciate my interest in this aspect, thank you!

Hello author, thank you for your excellent work and code sharing, I have a question, how to use this docker, how to place the source data and target data in the workspace directory, and which file to run to change the face