Image Classification on ImageNet (val) (Top-1 Accuracy and Computational Cost)
0.02Inference GFLOPSSTR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| STRSparsity (%)=97.78, Backbone=ResNet502021.06 | 0.02 | 62.84 | — | |
| STRSparsity (%)=94.8, Backbone=ResNet502021.06 | 0.04 | 70.97 | — | |
| Top-KASTSparsity (%)=98 fwd, 90 bwd, Backbone=ResNet502021.06 | 0.05 | 67.06 | 0.08 | |
| WoodFisherSparsity (%)=98, Backbone=ResNet502021.06 | 0.05 | 65.55 | — | |
| AC/DCSparsity (%)=98, Backbone=ResNet502021.06 | 0.06 | — | 0.46 | |
| RigLSparsity (%)=95, Backbone=ResNet502021.06 | 0.08 | — | 0.08 | |
| Top-KASTSparsity (%)=95 fwd, 50 bwd, Backbone=ResNet502021.06 | 0.08 | 71.96 | 0.22 | |
| WoodFisherSparsity (%)=95, Backbone=ResNet502021.06 | 0.09 | 72.12 | — | |
| AC/DCSparsity (%)=95, Backbone=ResNet502021.06 | 0.11 | — | 0.53 | |
| RigL (ERK)Sparsity (%)=95, Backbone=ResNet502021.06 | 0.12 | — | 0.13 | |
| DenseSparsity (%)=0, Backbone=ResNet502021.06 | 8.2 | 76.84 | 3.14 |