anguyen8 / deepface-emd Goto Github PK

Hi @anguyen8,
Thank your project.
I tried to convert resnet18_110.pth to onnx model, but the error:
"Unsupported: ONNX export of operator adaptive_avg_pool2d, output size that are not factor of input size. Please feel free to request support or submit a pull request on PyTorch GitHub."
Can you help me to solve this problem? Thank you.

I appreciate your insightful work, could you please release the code of LMK method?

Good job. I read the paper and code carefully, but I didn't find the training script. Can you provide some training script examples? thank you.

Hello!

Have you tried to use EMD with image features produced by visual transformer?
Looks like the last transformer layer of typical ViT outputs embeddings for each image patch, which is similar to the conv-features you use in the paper.

Is there any disadvantages to use such transformer features in your DeepFace-EMD method?
The only difference I see is that patch embeddings in ViT interact with each other throughout the whole image because of the attention mechanism, while conv-features are more local.

Thank you!

In the implementation for Spatial Correlation (emd_similarity function in emd.py), u and v are computed based on the distance instead of the similarity. Is it correct?

        elif method == 'sc':
            u = torch.sum(dis, 2)
            u = u / (u.sum(dim=1, keepdims=True) + 1e-7)
            v = torch.sum(dis, 1)
            v = v / (v.sum(dim=1, keepdims=True) + 1e-7)

Hello, Thanks for the great works!
I read the paper! I am wondering whether you have tested in arcface with resnet50 or larger networks? Would it still outperform other model without rank2? Thank you!

Meaningful work, when I want to use 128 * 128 to align people's faces, but I don't find ref_pts For 128x128? I haven't found the converted code, can you provide it?
thanks!