Image Classification on ImageNet (val) (FLOPs and Inference Speedup)
42.05FLOPs (%)HALP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HALPModel=ResNet50, variant=higher pruning, Finetune recipe=GBN2022.10 | 42.05 | 76.09 | — | — | |
| HALPModel=ResNet50, variant=higher pruning, Finetune recipe=QCQP2022.10 | 42.05 | 75.27 | — | — | |
| GBN-50Model=ResNet502022.10 | 44.94 | 75.18 | — | — | |
| HALPModel=ResNet50, Finetune recipe=ours2022.10 | 58.12 | 76.93 | 1.6 | -0.22 | |
| HALPModel=ResNet50, Finetune recipe=GBN2022.10 | 58.12 | 76.49 | — | — | |
| HALPModel=ResNet50, Finetune recipe=QCQP2022.10 | 58.12 | 76.53 | — | 0.16 | |
| QCQPModel=ResNet502022.10 | 59 | 76 | 1.52 | 0.32 | |
| GBN-60Model=ResNet502022.10 | 59.46 | 76.19 | — | — |