Image Classification on ImageNet 32 (val)
84.5Top-1 AccuracyFull-precision
Evaluation results
| Method | Links | |
|---|---|---|
| Full-precisionNetwork=ViT-B, #Bits=32/32, Size=346.32026.05 | 84.5 | |
| Full-precisionNetwork=Swin-S, #Bits=32/32, Size=198.42026.05 | 83.2 | |
| RepQ-ViT + NBCNetwork=Swin-S, #Bits=4/4, Size=33.72026.05 | 81.2 | |
| RepQ-ViT + QwT-v2Network=Swin-S, #Bits=4/4, Size=26.42026.05 | 80.3 | |
| RepQ-ViTNetwork=Swin-S, #Bits=4/4, Size=25.82026.05 | 80.2 | |
| RepQ-ViT + NBCNetwork=ViT-B, #Bits=4/4, Size=59.12026.05 | 77.6 | |
| RepQ-ViT + QwT-v2Network=ViT-B, #Bits=4/4, Size=45.62026.05 | 75.6 | |
| Full-precisionNetwork=DeiT-T, #Bits=32/32, Size=22.92026.05 | 72.1 | |
| RepQ-ViTNetwork=ViT-B, #Bits=4/4, Size=44.92026.05 | 69.1 | |
| RepQ-ViT + NBCNetwork=DeiT-T, #Bits=4/4, Size=4.22026.05 | 61.9 | |
| RepQ-ViT + QwT-v2Network=DeiT-T, #Bits=4/4, Size=3.42026.05 | 59.9 | |
| RepQ-ViTNetwork=DeiT-T, #Bits=4/4, Size=3.32026.05 | 57.6 |