FeatAug-DETR: Enriching One-to-Many Matching for DETRs with Feature Augmentation

Benchmark Model Rank Results
object-detection-on-coco-val2017H-Deformable-DETR + FeatAug-Flip (Swin-L, 24ep, 300 predictions)#45box AP: 58.3AP50: 77.1AP75: 64.0APS: 41.7APM: 62.4APL: 73.9
object-detection-on-coco-val2017Deformable-DETR w/ tricks + FeatAug-FC (Swin-L, 24ep, 300 predictions)#52box AP: 57.6AP50: 76.7AP75: 63.1APS: 41.1APM: 61.5APL: 73.8
object-detection-on-coco-val2017H-Deformable-DETR + FeatAug-Flip (ResNet-50, 24ep)#147box AP: 50.4AP50: 68.7AP75: 54.9APS: 32.7APM: 53.6APL: 65.2
object-detection-on-coco-val2017Deformable-DETR w/ tricks + FeatAug-FC (ResNet-50, 24ep)#156box AP: 49.9AP50: 68.3AP75: 54.7APS: 32.5APM: 52.9APL: 65.1