|
| 方法 | TOP-1准确率/+微调 | 计算量/计算量剪枝率 | 参数量/参数量剪枝率 |
| 原始ResNet56 |
| 94.40% | 0 | 0 |
| 传统的手工剪枝 | L1 | 94.20% | 27.56% | 14.12% |
| ThiNet50 [5] | 93.56% | 66.11% | 73.62% | |
| 基于AutoML的自动剪枝 | ABCpruner-70% | 93.55% | 56.61% | 57.14% |
| MetaPruning-0.5 [13] | 93.08% | 56.72% | 60.28% | |
| GAL-0.5 | 92.92%/93.10% | 48.52% | 44.76% | |
| GreedyPruner-0.5 | 91.83%/93.60% | 64.89% | 73.85% |