Network Pruning 벤치마크
Network Pruning on CIFAR-100
Accuracy
- 2019-05-23 — TAS-pruned ResNet-110: Accuracy 73.16
- 2021-06-23 — Dense: Accuracy 79.0
| Rank | Model | Accuracy | GFLOPs | Inference Time (ms) | Paper | Code | Year |
|---|---|---|---|---|---|---|---|
| 1 | Dense | 79 | – | – | AC/DC: Alternating Compressed/DeCompressed Training of Deep Neural Networks | IST-DASLab/ACDC · IST-DASLab/sparseprop | 2021 |
| 2 | AC/DC | 78.2 | – | – | AC/DC: Alternating Compressed/DeCompressed Training of Deep Neural Networks | IST-DASLab/ACDC · IST-DASLab/sparseprop | 2021 |
| 3 | Beta-Rank | 74.01 | – | – | Beta-Rank: A Robust Convolutional Filter Pruning Method For Imbalanced Medical Image Analysis | mohofar/beta-rank | 2023 |
| 4 | TAS-pruned ResNet-110 | 73.16 | 0.12 | – | Network Pruning via Transformable Architecture Search | D-X-Y/GDAS · D-X-Y/NAS-Projects · D-X-Y/AutoDL-Projects · +1 | 2019 |
| 5 | +U-DML* | – | – | 675.56 | PP-StructureV2: A Stronger Document Analysis System | PaddlePaddle/PaddleOCR | 2022 |