Image Classification on ImageNet 1K (Accuracy, FLOPs Ratio, and Latency Speedup Analysis)
65Accuracy DeltaHAWQ-V2
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| HAWQ-V2Model=ResNet-50, Joint P+Q=No (quant.), Search / Protocol=Hessian trace2026.06 | 65 | 13 | — | 76.1 | |
| DeepCompress-ViTModel=DeiT-S, Joint P+Q=No (storage + quant.), Search / Protocol=UCT + compres. training2026.06 | 50 | 14 | — | 79.8 | |
| I-ViTModel=DeiT-S, Joint P+Q=No (quant.), Search / Protocol=QAT (INT8)2026.06 | 27 | 4 | — | 79.85 | |
| AMCModel=MobileNet-V1, Joint P+Q=No (prune), Search / Protocol=RL (DDPG)2026.06 | 10 | 2 | — | 70.3 | |
| HAQModel=ResNet-50, Joint P+Q=No (quant.), Search / Protocol=RL (DDPG)2026.06 | 0 | — | 1.4 | 76.1 |