Image Classification on ImageNet (test) (Step-based Accuracy)
75.66ANN AccuracyLCP
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LCPArchitecture=ResNet-34 (with 2 extra layers in first conv-block), Evaluation Protocol=Post-Training2024.09 | 75.66 | 55.16 | 67.56 | 72.48 | 74.53 | 75.44 | |
| ACPArchitecture=ResNet-34 (with 2 extra layers in first conv-block), Evaluation Protocol=Training-involved2024.09 | 75.66 | 64.65 | 71.3 | 73.94 | 75 | 75.45 | |
| LCPArchitecture=MobileNet, Evaluation Protocol=Post-Training2024.09 | 73.4 | 0.2 | 15.47 | 55.95 | 66.35 | 72.19 | |
| Local threshold balancing (Ours)Architecture=ResNet-34, Evaluation Protocol=Post-Training2024.09 | 73.31 | 64.88 | 71.17 | 72.3 | 72.42 | 72.46 | |
| QCFSArchitecture=ResNet-34, Evaluation Protocol=Training-involved2024.09 | 73.3 | 64.43 | 71.51 | 73.25 | 73.67 | 73.76 | |
| Local threshold balancing (Ours*)Architecture=ResNet-34, Evaluation Protocol=Training-involved2024.09 | 73.3 | 71.94 | 73.06 | 73.5 | 73.49 | 73.54 | |
| Local threshold balancing (Ours)Architecture=MobileNet V2, Evaluation Protocol=Post-Training2024.09 | 71.88 | 19.39 | 49.93 | 64.31 | 68.89 | 70.48 | |
| RMPArchitecture=ResNet-34 (without BN), Evaluation Protocol=Post-Training2024.09 | 70.64 | — | — | 55.65 | 60.08 | — | |
| Local threshold balancing (Ours)Architecture=ResNet-18, Evaluation Protocol=Post-Training2024.09 | 69.76 | 65.55 | 68.53 | 69.03 | 69.21 | 69.21 |