Object Detection on SODA-A
83.8AIUHR-DETR
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| UHR-DETRBackbone=ResNet-502026.04 | 83.8 | 53.4 | 45.9 | 20.8 | 56.8 | 36.3 | 69.4 | 51.4 | 61.9 | 53.3 | 30.9 | 59.4 | 40.8 | 0.647 | 5.57 | |
| CEASCBackbone=ResNet-502026.04 | 81.6 | 61.4 | 45.7 | 37.6 | 63.6 | 43 | 66.5 | 80.2 | 56.1 | 59.5 | 25.7 | 69.7 | 48.3 | 4.467 | 5.38 | |
| SparseFormerBackbone=SparseNet, Training protocol=from scratch for 36 epochs following [34]2026.04 | 80 | 62.6 | 47.1 | 37.3 | 61.4 | 46.8 | 63.3 | 83.1 | 40.6 | 58 | 24 | 68 | 53.6 | 4.802 | 2.51 | |
| GigaDetBackbone=ResNet-502026.04 | 68.8 | 16.1 | 45.9 | 10.5 | 66.5 | 31.2 | 45 | 21.4 | 13.6 | 35.5 | 23.5 | 40.2 | 25.6 | 0.588 | 6.72 | |
| SPDetBackbone=CSPDarknet2026.04 | 64.5 | 8.2 | 31.9 | 20.9 | 42.8 | 38.8 | 55.1 | 70.8 | 18 | 39 | 13.1 | 48.5 | 39.3 | 1.957 | 0.75 | |
| RT-DETRBackbone=ResNet-50, Downsampling=1024x10242026.04 | 29 | 23.4 | 2.3 | 4.8 | 8.1 | 19 | 11.1 | 48 | 19 | 18.3 | 1.2 | 21 | 34.3 | 0.036 | 0.46 | |
| FoveaBackbone=ResNet-50, Downsampling=1024x10242026.04 | 5 | 0 | 0 | 2.5 | 1 | 1 | 3.5 | 2.8 | 1.5 | 1.9 | 0.3 | 2 | 4.6 | 2.1 | 2.53 |