I’m start to use Pytorch and trying to implement ResNet for my problem, which is not related to images or NLP. The implementation of ResNet in Pytorch has convolutional layers which I don’t need and don’t want to include at the beginning. Are you guys aware of existing implementation of ResNet in Pytorch in a MLP type of setting? Or what I really need to do is reimplement the ResNet in Pytorch?
If you are removing the conv layers in the ResNet implementation, I guess you would like to create a fully connected model with some skip connections?
If so, you could have a look at the current implementation of BasicBlock and BottleNeck and try to adapt these modules to your code.
Thank you. That’s actually what I’m doing currently. I try read and modify the code in github: pytorch/vision/blob/master/torchvision/models/resnet.py.