Reproduce “Isomorphic Pruning for Vision Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Gongfan Fang, Xinyin Ma, Michael Bi Mi, Xinchao Wang
Isomorphic pruning makes image-classification models smaller by removing coupled dimensions and ranking only structures with matching computational roles, rather than comparing unlike network parts. On ImageNet-1K, a one-shot-pruned DeiT-T, fine-tuned for 300 epochs, reaches 77.50% top-1 accuracy with 1.21 G MACs, versus 74.52% and 1.27 G for standard DeiT-T; fewer MACs do not establish faster execution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.