I want to ask whether there is a method to prune channels of trained ResNet model while preserving the unpruned channels’ weight? The pruning method I used is similar to "Pruning from Scratch" paper, but I want to use my trained ResNet-50 instead of training from scratch.
I think of whether there's a method to load the trained model's checkpoint and perform channel-wise pruning, so the unpruned channels will have it's original weight preserved.