Oriented Object Detection on FAIR-1M 2.0
47.4AP50RiO-DETR-x
Evaluation results
| Method | Links | |
|---|---|---|
| RiO-DETR-xBackbone=HGNet-B5, #P=62.5M, Flops=527G, Lat.=29.9, multi-scale training and testing protocol=true2026.03 | 47.4 | |
| Strip-RCNN-SBackbone=StripNet-S, #P=30.5M, Flops=159G, Lat.=241.9, multi-scale training and testing protocol=true2026.03 | 46.8 | |
| YOLO26x-obbBackbone=YOLO26x, #P=57.6M, Flops=517G, Lat.=30.5, multi-scale training and testing protocol=true2026.03 | 46.7 | |
| LSKNet-SBackbone=LSKNet-S, #P=31.0M, Flops=161G, Lat.=203.5, multi-scale training and testing protocol=true2026.03 | 46.3 | |
| LOOD (RD)Backbone=R-101, multi-scale training and testing protocol=true2026.03 | 44.9 | |
| PKINet-SBackbone=PKINet-S, #P=30.8M, Flops=190G, Lat.=359.7, multi-scale training and testing protocol=true2026.03 | 44.5 | |
| RiO-DETR-mBackbone=HGNet-B2, #P=18.6M, Flops=158G, Lat.=8.8, multi-scale training and testing protocol=true2026.03 | 43.6 | |
| ReDetBackbone=ReR-50, #P=31.8M, Flops=225G, Lat.=448.2, multi-scale training and testing protocol=true2026.03 | 43.2 | |
| LOOD (RT)Backbone=R-50, multi-scale training and testing protocol=true2026.03 | 42.6 | |
| YOLO26m-obbBackbone=YOLO26m, #P=21.2M, Flops=183G, Lat.=10.2, multi-scale training and testing protocol=true2026.03 | 42.5 | |
| O-RCNNBackbone=R-101, #P=60.3M, Flops=289G, Lat.=160.8, multi-scale training and testing protocol=true2026.03 | 40.4 | |
| RoI Trans.Backbone=R-101, #P=67.8M, Flops=607G, multi-scale training and testing protocol=true2026.03 | 40.2 | |
| O-RepPointsBackbone=Swin-T, #P=37.3M, Flops=200G, multi-scale training and testing protocol=true2026.03 | 38.9 | |
| R-Faster RCNNBackbone=R-101, #P=60.1M, Flops=289G, Lat.=163.4, multi-scale training and testing protocol=true2026.03 | 37.5 | |
| S2A-NetBackbone=R-50, #P=31.6M, Flops=588G, multi-scale training and testing protocol=true2026.03 | 37.4 | |
| R-FCOSBackbone=R-101, #P=50.9M, Flops=284G, Lat.=156.9, multi-scale training and testing protocol=true2026.03 | 36.1 | |
| SASM RepPointsBackbone=R-101, #P=55.8M, Flops=542G, Lat.=144.5, multi-scale training and testing protocol=true2026.03 | 30.9 |