Object Detection on MS-COCO (Base/Novel AP Metrics)
50.3APS2D
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| S2DPrecision=FP162026.02 | 50.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdamWPrecision=FP162026.02 | 50.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| S2DQuantization Method=ERQ, Bit-width (W/A)=W8A82026.02 | 49.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdamWQuantization Method=ERQ, Bit-width (W/A)=W8A82026.02 | 48.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| S2DQuantization Method=PTQ4ViT, Bit-width (W/A)=W8A82026.02 | 48.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| S2DQuantization Method=ERQ, Bit-width (W/A)=W6A62026.02 | 47.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdamWQuantization Method=ERQ, Bit-width (W/A)=W6A62026.02 | 45.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdamWQuantization Method=PTQ4ViT, Bit-width (W/A)=W8A82026.02 | 42.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| S2DQuantization Method=ERQ, Bit-width (W/A)=W5A52026.02 | 40.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Decoupling Classifiershots=302025.05 | 32.3 | 51.3 | 35.8 | — | 21.9 | 40.2 | — | — | — | — | — | — | 34.5 | |
| Decoupling Classifiershots=102025.05 | 31.4 | 49.9 | 35.7 | — | 18.3 | 34.5 | — | — | — | — | — | — | 33.4 | |
| DeFRCNshots=302025.05 | 31.4 | 48.8 | 34.8 | — | 21.2 | 36.7 | — | — | — | — | — | — | 33.9 | |
| Decoupling Classifiershots=52025.05 | 30.2 | 48.2 | 35 | — | 15.7 | 30.3 | — | — | — | — | — | — | 32.2 | |
| DeFRCNshots=102025.05 | 29.7 | 46 | 34 | — | 16.8 | 29.6 | — | — | — | — | — | — | 32.1 | |
| Decoupling Classifiershots=32025.05 | 29.4 | 46.8 | 34.9 | — | 12.9 | 25.1 | — | — | — | — | — | — | 31.4 | |
| TFAshots=302025.05 | 28.7 | 44.7 | 34.2 | — | 12.1 | 22 | — | — | — | — | — | — | 31.5 | |
| Decoupling Classifiershots=22025.05 | 28.6 | 45.6 | 34.7 | — | 10.4 | 20.9 | — | — | — | — | — | — | 30.7 | |
| DeFRCNshots=52025.05 | 27.8 | 43 | 32.6 | — | 13.6 | 24.7 | — | — | — | — | — | — | 30.2 | |
| Decoupling Classifiershots=12025.05 | 27.4 | 43.4 | 34.4 | — | 6.2 | 12.7 | — | — | — | — | — | — | 29.4 | |
| DeFRCNshots=32025.05 | 26.6 | 41.1 | 32.1 | — | 10.7 | 20 | — | — | — | — | — | — | 28.9 | |
| TFAshots=102025.05 | 26.6 | 42.2 | 32.4 | — | 9.1 | 17.1 | — | — | — | — | — | — | 29 | |
| TFAshots=52025.05 | 25.9 | 41.2 | 32.3 | — | 7 | 13.3 | — | — | — | — | — | — | 28.4 | |
| DeFRCNshots=22025.05 | 25.7 | 39.6 | 31.4 | — | 8.5 | 16.3 | — | — | — | — | — | — | 28 | |
| TFAshots=32025.05 | 25.3 | 40.4 | 32 | — | 5.1 | 9.9 | — | — | — | — | — | — | 27.6 | |
| TFAshots=22025.05 | 24.9 | 40.1 | 31.9 | — | 3.9 | 7.8 | — | — | — | — | — | — | 27 | |
| S2DQuantization Method=PTQ4ViT, Bit-width (W/A)=W6A62026.02 | 24.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| TFAshots=12025.05 | 24.4 | 39.8 | 31.9 | — | 1.9 | 3.8 | — | — | — | — | — | — | 26.1 | |
| DeFRCNshots=12025.05 | 24 | 36.9 | 30.4 | — | 4.8 | 9.5 | — | — | — | — | — | — | 26.2 | |
| S2DQuantization Method=ERQ, Bit-width (W/A)=W4A42026.02 | 16.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FRCN+ftshots=12025.05 | 16.2 | 25.8 | 21 | — | 1.7 | 3.3 | — | — | — | — | — | — | 17.6 | |
| FRCN+ftshots=22025.05 | 15.8 | 25 | 20 | — | 3.1 | 6.1 | — | — | — | — | — | — | 17.3 | |
| FRCN+ftshots=32025.05 | 15 | 23.9 | 18.8 | — | 3.7 | 7.1 | — | — | — | — | — | — | 16.4 | |
| FRCN+ftshots=52025.05 | 14.4 | 23 | 17.6 | — | 4.6 | 8.7 | — | — | — | — | — | — | 15.6 | |
| FRCN+ftshots=302025.05 | 13.5 | 21.8 | 15.6 | — | 7.4 | 13.1 | — | — | — | — | — | — | 14.5 | |
| FRCN+ftshots=102025.05 | 13.4 | 21.8 | 16.1 | — | 5.5 | 10 | — | — | — | — | — | — | 14.5 | |
| AdamWQuantization Method=ERQ, Bit-width (W/A)=W5A52026.02 | 10.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdamWQuantization Method=PTQ4ViT, Bit-width (W/A)=W6A62026.02 | 2.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdamWQuantization Method=ERQ, Bit-width (W/A)=W4A42026.02 | 1.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Li et al.Initialization=[53], Backbone=ResNet-50, Detector=Faster R-CNN, AT recipe=Li et al. [53]2026.01 | — | — | — | — | — | — | 29.9 | 49.3 | 14.8 | 25.5 | 19.7 | 40.5 | — | |
| MPMInitialization=MPM, Backbone=ResNet-50, Detector=Faster R-CNN, AT recipe=Li et al. [53]2026.01 | — | — | — | — | — | — | 30.3 | 50.3 | 14.8 | 25.9 | 19.8 | 40.7 | — | |
| Vanilla BaselineInitialization=vanilla, Backbone=ResNet-50, Detector=Faster R-CNN, AT recipe=Li et al. [53]2026.01 | — | — | — | — | — | — | 29.9 | 49.8 | 14.5 | 25.4 | 19.3 | 40.2 | — |