MSVFP
PyTorch checkpoints for Magnitude and Similarity Based Variable Rate Filter Pruning for Efficient Convolution Neural Networks, by Deepak Ghimire and Seong-Heum Kim. See the published article and the original code and evaluation instructions.
Checkpoints
The VGG16 pair is the first of the three CIFAR-10 runs used by the study. The ImageNet files include a dense baseline and both available pruned variants for each ResNet architecture. The pruning target labels come from the original checkpoint folders. source-checkpoints.json records the original Git paths and blob IDs.
| Dataset | Architecture | Dense checkpoint | Pruned checkpoints |
|---|---|---|---|
| CIFAR-10 | VGG16 | Round 1 | 34.2% target |
| ImageNet | ResNet18 | Dense | 41% target, 45% target |
| ImageNet | ResNet34 | Dense | 41% target, 45% target |
| ImageNet | ResNet50 | Dense | 42% target, 53% target |
Download and evaluate
Install the Hub client with pip install huggingface_hub, then download the VGG16 pair.
hf download ghimiredhikura/MSVFP --include "CIFAR-10/*" --local-dir checkpoints
From the project checkout, follow the matching evaluation command in the GitHub README. Set --pretrain_path to the dense checkpoint and --pruned_path to the corresponding pruned checkpoint. The files retain their original PyTorch serialization. Use the project's Python/PyTorch environment and load checkpoints from trusted sources.
Usage terms and citation
The source repository provides no explicit license for these checkpoint files. Contact the authors for reuse or redistribution terms. Cite the paper when using this work.