Object Detection on MS COCO 2017 (minival)
46.7APD-DETR+Ours
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| D-DETR+Oursnumber of queries=3002022.03 | 46.7 | 65.3 | 50.3 | 28.6 | 49.8 | 61.7 | |
| S-RCNN+Oursnumber of queries=3002022.03 | 46.1 | 65.3 | 50.6 | 29.2 | 48.7 | 59.9 | |
| D-DETRnumber of queries=3002022.03 | 45.8 | 64.5 | 49.4 | 28.2 | 49 | 61.7 | |
| AdaMixerBackbone=ResNet-50, Epochs=12, Number of queries (N)=5002022.03 | 45 | 64.2 | 48.6 | 27.9 | 47.8 | 61.1 | |
| S-RCNNnumber of queries=3002022.03 | 45 | 64.2 | 49.1 | 27.6 | 47.5 | 59.1 | |
| AdaMixerBackbone=ResNet-50, Epochs=12, Number of queries (N)=3002022.03 | 44.1 | 63.4 | 47.4 | 27 | 46.9 | 59.5 | |
| AdaMixerBackbone=ResNet-50, Epochs=12, Number of queries (N)=1002022.03 | 42.7 | 61.5 | 45.9 | 24.7 | 45.4 | 59.2 | |
| Dynamic HeadBackbone=ResNet-50, Epochs=122022.03 | 42.6 | 60.1 | 46.4 | 26.1 | 46.8 | 56 | |
| R3-CNN (advanced)# Params=50,072 K, TS=Inc, Lt=3, H=1, Speed=1.0, Mem. usage=6.8 GB, Model size=384 MB2021.04 | 42 | — | — | — | — | — | |
| HTC# Params=77,230 K, TS=Inc, Lt=3, H=3, Speed=5.4, Mem. usage=6.8 GB, Model size=591 MB2021.04 | 41.7 | — | — | — | — | — | |
| BorderDetBackbone=ResNet-50, Epochs=122022.03 | 41.4 | 59.4 | 44.5 | 23.6 | 45.1 | 54.6 | |
| GFocalV2Backbone=ResNet-50, Epochs=122022.03 | 41.1 | 58.8 | 44.9 | 23.5 | 44.9 | 53.3 | |
| R3-CNN (naive)# Params=43,912 K, TS=Inc, Lt=3, H=1, Speed=5.5, Mem. usage=5.9 GB, Model size=337 MB2021.04 | 40.9 | — | — | — | — | — | |
| Cascade R-CNNBackbone=ResNet-50, Epochs=122022.03 | 40.4 | 58.9 | 44.1 | 22.8 | 43.7 | 54 | |
| R³-CNN (naive)# Params=43,912 K, TS=Dec, Lt=3, H=1, Speed=5.5, Mem. usage=5.9 GB, Model size=337 MB2021.04 | 40.4 | — | — | — | — | — | |
| GC-net + GROIEBackbone=r50-FPN2020.04 | 40.3 | 62.4 | 44 | 24.2 | 44.4 | 52.5 | |
| Upper BoundScenario=Full data, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 40.2 | 58.3 | 43.6 | 23.2 | 44.1 | 52.2 | |
| Grid R-CNN + GROIEBackbone=r50-FPN2020.04 | 39.8 | 58.1 | 42.9 | 23.6 | 43.9 | 51.5 | |
| GC-netBackbone=r50-FPN2020.04 | 39.5 | 62 | 42.7 | 24.6 | 43.2 | 51.6 | |
| Mask (3x)# Params=44,170 K, TS=-, Lt=1, H=1, Speed=5.4, Mem. usage=4.4 GB, Model size=339 MB2021.04 | 39.2 | — | — | — | — | — | |
| Grid R-CNNBackbone=r50-FPN2020.04 | 39.1 | 57.2 | 42.2 | 22.1 | 43 | 50.6 | |
| FCOSBackbone=ResNet-50, Epochs=122022.03 | 38.7 | 57.4 | 41.8 | 22.9 | 42.5 | 50.1 | |
| Mask R-CNN + GROIEBackbone=r50-FPN2020.04 | 38.4 | 59.9 | 41.7 | 22.9 | 42.1 | 49.7 | |
| Mask (1x)# Params=44,170 K, TS=-, Lt=1, H=1, Speed=11.5, Mem. usage=4.4 GB, Model size=339 MB2021.04 | 38.2 | — | — | — | — | — | |
| Sparse R-CNNBackbone=ResNet-50, Epochs=122022.03 | 37.9 | 56 | 40.5 | 20.7 | 40 | 53.5 | |
| Faster R-CNN + GROIEBackbone=r50-FPN2020.04 | 37.5 | 59.2 | 40.6 | 22.3 | 41.5 | 47.8 | |
| Mask R-CNNBackbone=r50-FPN2020.04 | 37.3 | 58.9 | 40.4 | 21.7 | 41.1 | 48.2 | |
| ERDScenario=40 classes + 40 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 36.9 | 54.5 | 39.6 | 21.3 | 40.4 | 47.5 | |
| ERDScenario=50 classes + 30 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 36.6 | 54 | 38.9 | 19.4 | 40.4 | 48 | |
| Faster R-CNNBackbone=r50-FPN2020.04 | 36.5 | 58.4 | 39.1 | 21.9 | 40.4 | 46.8 | |
| ERDScenario=60 classes + 20 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 35.8 | 52.9 | 38.4 | 20.6 | 39.4 | 46.5 | |
| Deformable DETRBackbone=ResNet-50, Epochs=122022.03 | 35.1 | 53.6 | 37.7 | 18.2 | 38.5 | 48.7 | |
| ERDScenario=70 classes + 10 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 34.9 | 51.9 | 37.4 | 18.7 | 38.8 | 45.5 | |
| SIDScenario=40 classes + 40 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 34 | 51.4 | 36.3 | 18.4 | 38.4 | 44.9 | |
| SIDScenario=50 classes + 30 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 33.8 | 51 | 36.1 | 17.6 | 38.1 | 45.1 | |
| SIDScenario=70 classes + 10 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 32.8 | 49 | 35 | 17.1 | 36.9 | 44.5 | |
| SIDScenario=60 classes + 20 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 32.7 | 49.8 | 34.6 | 17.2 | 37.6 | 43.5 | |
| RILODScenario=40 classes + 40 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 29.9 | 45 | 32 | 15.8 | 33 | 40.5 | |
| RILODScenario=50 classes + 30 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 28.5 | 43.2 | 30.2 | 15.4 | 31.6 | 38 | |
| RILODScenario=60 classes + 20 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 25.4 | 38.8 | 26.8 | 13.9 | 29 | 33.7 | |
| RILODScenario=70 classes + 10 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 24.5 | 37.9 | 25.7 | 14.2 | 27.4 | 33.5 | |
| DETRBackbone=ResNet-50, Epochs=122022.03 | 20 | 36.2 | 19.3 | 6 | 20.5 | 32.2 | |
| Catastrophic ForgettingScenario=40 classes + 40 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 17.8 | 25.9 | 19.3 | 8.3 | 19.2 | 24.6 | |
| LwFScenario=40 classes + 40 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 17.2 | 25.4 | 18.6 | 7.9 | 18.4 | 24.3 | |
| Catastrophic ForgettingScenario=50 classes + 30 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 14.1 | 20.6 | 15.2 | 7 | 14.5 | 19.2 | |
| Catastrophic ForgettingScenario=60 classes + 20 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 9.8 | 14 | 10.6 | 4.3 | 14.1 | 13.5 | |
| LwFScenario=70 classes + 10 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 7.1 | 12.4 | 7 | 4.8 | 9.5 | 10 | |
| LwFScenario=60 classes + 20 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 5.8 | 10.8 | 5.3 | 4 | 8.5 | 7.7 | |
| LwFScenario=50 classes + 30 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 5 | 9.5 | 4.6 | 5 | 6.7 | 5.7 | |
| Catastrophic ForgettingScenario=70 classes + 10 classes, Detector=GFLV1, Backbone=ResNet-50, Neck=FPN2022.04 | 4.3 | 6.5 | 4.5 | 2.1 | 5.1 | 6.8 |