search

qubvel / classification_models

Classification models trained on ImageNet. Keras.
MIT License
1.22k stars 307 forks source link

Resnet models #28

Closed tinalegre closed 5 years ago

tinalegre commented 5 years ago

It looks that your own RestNet implementation is based on Resnet Version 2 (Preact / BN->Relu->Conv). Could you please also consider adding ResNet Version 1 (code + pre-trained weights)? I'm unfortunately limited on GPU resources and can't therefore do a full train in imagenet, etc. I know that keras-applications has both V1/V2, but I find that your toolbox is better organized.

qubvel commented 5 years ago

Hi @tinalegre I did not train any of models, just convert or transfer weights from other frameworks/repos. If you will find such pretrained models I will consider such option. Models from keras_applications are inside of this repo.

tinalegre commented 5 years ago

Hi @qubvel thank you. I was wondering why on your ResNet model, just after the last residual block and before the top classification layer, we need to have the additional BN+Relu layers (named respectively bn1 and relu1)?

qubvel commented 5 years ago

Once again: ResNet models was not created by me. I have just rewrite architecture from MXNet models zoo as it is and convert weights. So the answer to your question is: "Because they have been created and trained with such architecture by someone (I actually don`t know the authors of this implementation and their motivation)"