search

clovaai / wsolevaluation

Evaluating Weakly Supervised Object Localization Methods Right (CVPR 2020)
MIT License
332 stars 55 forks source link

Pretrained Resnet #37

Closed shadowwkl closed 3 years ago

shadowwkl commented 4 years ago

Hi,

I notice that for ResNet-50, you change the stride to 1 at Layer 3 (the name comes from pytorch, torchvision.models.resnet50), in order to increase the feature map from 7x7 to 14x14. So I wonder do you first do this change and then use ImageNet to train the modified ResNet-50, and finally based on this trained version new ResNet-50, you train (or finetune) it using CUB dataset?

junsukchoe commented 4 years ago

Hello,

We did not pre-train our customized networks and just use ImageNet pre-trained models from PyTorch. I hope that this reply is helpful!

shadowwkl commented 4 years ago

Hello,

We did not pre-train our customized networks and just use ImageNet pre-trained models from PyTorch. I hope that this reply is helpful!

Hi!

Thanks for your reply. So you keep all the weights from the imagenet-pretrained model except the two last strided conv layers, which originally have stride 2 and now are set to 1 in order to increase the feature map from 7x7 to 14x14. Do I understand correctly?

junsukchoe commented 3 years ago

So you keep all the weights from the imagenet-pretrained model except the two last strided conv layers,

No, we use all weights of ImageNet pretrained model, including the last two strided conv layers.

shadowwkl commented 3 years ago

So you keep all the weights from the imagenet-pretrained model except the two last strided conv layers,

No, we use all weights of ImageNet pretrained model, including the last two strided conv layers.

Thank you! Problem solved.