Image Classification on ImageNet-1K (val) (Accuracy, Latency, GFLOPS)
87.1AccuracyViT-H+ Uniform
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ViT-H+ UniformNumber of fixations=1, Sampling resolution=224, Pixels=50176, Patches=1962026.02 | 87.1 | 445.16 | 172.36 | |
| FOVI-ViT-H+Number of fixations=3, Sampling resolution=64, a parameter=2.79, Pixels=3976, Patches=642026.02 | 85 | 475.7 | 175.26 | |
| FOVI-ViT-H+Number of fixations=1, Sampling resolution=64, a parameter=2.79, Pixels=3976, Patches=642026.02 | 84 | 171.53 | 58.42 | |
| ViT-S+ UniformNumber of fixations=1, Sampling resolution=224, Pixels=50176, Patches=1962026.02 | 79.4 | 50.18 | 6.16 | |
| FOVI-ViT-S+Number of fixations=3, Sampling resolution=128, a parameter=2.79, Pixels=16078, Patches=642026.02 | 76 | 76.1 | 6.16 | |
| FOVI-ViT-S+Number of fixations=3, Sampling resolution=64, a parameter=2.79, Pixels=3976, Patches=642026.02 | 73.5 | 74.04 | 6.11 | |
| FOVI-ViT-S+Number of fixations=1, Sampling resolution=128, a parameter=2.79, Pixels=16078, Patches=642026.02 | 72.9 | 26.43 | 2.05 | |
| Weak FOVI-ViT-S+Number of fixations=3, Sampling resolution=64, a parameter=60.94, Pixels=4032, Patches=642026.02 | 72.5 | 74.11 | 6.12 | |
| ViT-S+ UniformNumber of fixations=3, Sampling resolution=64, Pixels=4096, Patches=642026.02 | 71.1 | 68.07 | 6.05 | |
| FOVI-ViT-S+Number of fixations=1, Sampling resolution=64, a parameter=2.79, Pixels=3976, Patches=642026.02 | 70 | 26.36 | 2.04 | |
| Weak FOVI-ViT-S+Number of fixations=1, Sampling resolution=64, a parameter=60.94, Pixels=4032, Patches=642026.02 | 69.3 | 26.32 | 2.04 | |
| ViT-S+ UniformNumber of fixations=1, Sampling resolution=64, Pixels=4096, Patches=642026.02 | 67.2 | 24.5 | 2.02 | |
| ViT-S+ Log-polarNumber of fixations=3, Sampling resolution=64, a parameter=2.79, Pixels=4096, Patches=642026.02 | 66.4 | 68.07 | 6.05 | |
| ViT-S+ Log-polarNumber of fixations=1, Sampling resolution=64, a parameter=2.79, Pixels=4096, Patches=642026.02 | 61.4 | 24.5 | 2.02 |