Table 6.
Comparison of different models on Params, FLOPs and FPS.
| Algorithms | Params(M) | FLOPs(M) | FPS |
|---|---|---|---|
| CLIP | 102(↑44.4) | 14800(↑3600) | 5.8(↓24.2) |
| PaLI | 17000(↑16942.4) | 25000(↑13800) | 20(↓10.0) |
| CoAtNet-7 | 24400(↑24342.4) | 12500(↑1300) | 50(↑12.0) |
| Classical ViT | 57.6 | 11200 | 30 |
| Improved ViT | 48.0(↓9.60) | 8800(↓2400) | 38 (↑8.0) |
| Improved ViT+KAN | 49.0(↓8.60) | 9000(↓2200) | 39 (↑9.0) |
| Improved ViT+BiFormer | 47.3(↓10.3) | 8333(↓2867) | 42 (↑12) |
| Improved ViT+KAN+BiFormer | 48.6(↓9.00) | 8533(↓3067) | 44 (↑14) |
The arrow pointing upwards “↑” indicates an increase, while the arrow pointing upwards “↓” indicates a decrease.