A PyTorch based DCNN channel/filter pruning implementation with virtual pruning. The pruned weights/features are ineffective during training/inference, which mimic the influence of pruning on accuracy/loss. But no real inference speed-up is introduced. This repo is released for research publication. More improvement and readability are coming...
scott-mao/pseudopruner
A PyTorch based DCNN channel/filter pruning implementation with virtual pruning. Realistic inference effect, no real speed-up.
Python