About the pre-trained eca_mobilenetv2 model.

[Kurumi233]

When I load the pre-trained model, some paramters are not match, and I change 96 in following code to 64, it fixed.

Is it a bug of model or paramters ?

https://github.com/BangguWu/ECANet/blob/cf8a4c8b3d49b27c12e98ceb930d0f7db6c2460a/models/eca_mobilenetv2.py#L77

[amdslgl]

hello,sir.Extracting the model tar diretory has an error ,do you have the same question?

[Kurumi233]

hello,sir.Extracting the model tar diretory has an error ,do you have the same question?

No, I just have an error I metioned above.

Are you extracting the tar file to load the model? Just torch.load('***.tar') will be ok.

[amdslgl]

thank you,sir.I have solve my question from yours.maybe ,your question can be handled by this: model = ResNet(ECABottleneck, [3, 4, 6, 3], num_classes=num_classes, k_size=k_size) if pretrained:

pretrained_model = model_zoo.load_url(model_urls['resnet50'])

    pretrained_model=torch.load('/home/lgl/.cache/torch/checkpoints/eca_resnet50_k3557.pth.tar')
    state = model.state_dict()
    for key in state.keys():
        if key in pretrained_model.keys():
            state[key] = pretrained_model[key]
    model.load_state_dict(state)

[Kurumi233]

thank you,sir.I have solve my question from yours.maybe ,your question can be handled by this: model = ResNet(ECABottleneck, [3, 4, 6, 3], num_classes=num_classes, k_size=k_size) if pretrained:

pretrained_model = model_zoo.load_url(model_urls['resnet50'])

pretrained_model=torch.load('/home/lgl/.cache/torch/checkpoints/eca_resnet50_k3557.pth.tar') state = model.state_dict() for key in state.keys(): if key in pretrained_model.keys(): state[key] = pretrained_model[key] model.load_state_dict(state)

Yes, thanks. I just want to use a totally pretrained model, so I have to change the model file.

[lph529372693]

@BangguWu @amdslgl @Kurumi233 Hello! Thank you for your repo and answers! Your method helps me solve the problem that the parameters do not match the model.But when ECA_ Mobilenetv2 pretraining was loaded into my model, during the training of my model, it was found that MIOU was only 3.5%, and almost no growth. Why is this? PS：Part of my mobilnet network uses serial atrous convolution and ECA block in your code.