Other ConvNets as the Backbone

[ahkarami]

Dear @layumi, Have you ever trained other ConvNets (e.g., ResNet101, ResNet152, ResNext, Inception-ResNet-V2) as the backbone of your code?

[layumi]

Hi @ahkarami I haven't trained other convnets. But I think this code can be converted to trian them. You mainly need to change the code in the model.py.

[ahkarami]

Dear @layumi, Thank you for your answer. I also appreciate you for adding DenseNet-121 to model.py. I encourage you to add other models (e.g., ResNext, DualPathNet, Inception-ResNet-V2, ...). You can use the below link for other models: Pre-Trained Models on PyTorch At last, it would be great if you release the weights of your trained models (e.g., DenseNet-121). Thank you very much.

[layumi]

Thank you for your suggestions, @ahkarami For the time being, I may not add more models. The main reason is that ResNet-50 is a common choice in most academic papers. I want to keep this repos simple and it can be in a fair comparison with other methods.

P.S. The Dense121 model can be found at Here . I just uploaded it. Many thanks.

[ahkarami]

Dear @layumi, Thank you very much for your complete answer. Yes, I agree with you there. ResNet50 is a common choice and great as a baseline model.