Object Detection on MS COCO novel classes val5k (minival)
16.6APAttention-RPN / Multi-Relation Detector
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Attention-RPN / Multi-Relation Detectordataset=FSOD+, fine-tune=None2019.08 | 16.6 | 31.3 | 16.1 | — | — | — | — | — | — | — | — | — | |
| MPSRShot=302020.07 | 14.1 | 25.4 | 14.2 | 400 | 1,290 | 23 | 17.7 | 24.2 | 24.3 | 550 | 21 | 39.3 | |
| Baseline-FPNShot=302020.07 | 13.7 | 25.1 | 13.3 | 360 | 1,250 | 23.3 | 17.8 | 24.7 | 24.7 | 540 | 21.6 | 40.5 | |
| BaselineShot=302020.07 | 12.6 | 25.7 | 11 | 320 | 1,180 | 20.7 | 15.9 | 21.8 | 21.8 | 510 | 18 | 36.9 | |
| Meta R-CNNShot=30, Pipeline=FRCN pipeline, Backbone=ResNet-502019.09 | 12.4 | 25.3 | 10.8 | 280 | 1,160 | 1,900 | 15 | 21.4 | 21.7 | 860 | 2,000 | 3,210 | |
| Meta R-CNNShot=302020.07 | 12.4 | 25.3 | 10.8 | 280 | 1,160 | 19 | 15 | 21.4 | 21.7 | 860 | 20 | 32.1 | |
| FRCN+ft-fullShot=30, Pipeline=FRCN pipeline, Backbone=ResNet-502019.09 | 11.1 | 21.6 | 10.3 | 290 | 880 | 1,890 | 15 | 21.1 | 21.3 | 1,010 | 1,790 | 3,320 | |
| Attention-RPN / Multi-Relation Detectordataset=COCO, fine-tune=MS COCO2019.08 | 11.1 | 20.4 | 10.6 | — | — | — | — | — | — | — | — | — | |
| MPSRShot=102020.07 | 9.8 | 17.9 | 9.7 | 330 | 920 | 16.1 | 15.7 | 21.2 | 21.2 | 460 | 19.6 | 34.3 | |
| Baseline-FPNShot=102020.07 | 9.5 | 17.3 | 9.4 | 270 | 840 | 15.9 | 14.8 | 20.6 | 20.6 | 470 | 19.3 | 33.1 | |
| YOLO-Few-shotShot=30, Pipeline=FRCN pipeline, Backbone=ResNet-502019.09 | 9.1 | 19 | 7.6 | 80 | 490 | 1,680 | 13.2 | 17.7 | 17.8 | 150 | 1,040 | 3,350 | |
| YOLO-FSShot=302020.07 | 9.1 | 19 | 7.6 | 80 | 490 | 16.8 | 13.2 | 17.7 | 17.8 | 150 | 10.4 | 33.5 | |
| BaselineShot=102020.07 | 8.8 | 18.7 | 7.1 | 290 | 810 | 15 | 12.9 | 17.2 | 17.2 | 410 | 14.2 | 29.1 | |
| Meta R-CNNShot=10, Pipeline=FRCN pipeline, Backbone=ResNet-502019.09 | 8.7 | 19.1 | 6.6 | 230 | 770 | 1,400 | 12.6 | 17.8 | 17.9 | 780 | 1,560 | 2,720 | |
| Meta R-CNNdataset=COCO, fine-tune=MS COCO2019.08 | 8.7 | 19.1 | 6.6 | — | — | — | — | — | — | — | — | — | |
| Meta R-CNNShot=102020.07 | 8.7 | 19.1 | 6.6 | 230 | 770 | 14 | 12.6 | 17.8 | 17.9 | 780 | 15.6 | 27.2 | |
| FRCN+ft-fullShot=10, Pipeline=FRCN pipeline, Backbone=ResNet-502019.09 | 6.5 | 13.4 | 5.9 | 180 | 530 | 1,130 | 12.6 | 17.7 | 17.8 | 650 | 1,440 | 2,860 | |
| Meta R-CNNshot=20, Backbone=ResNet-502019.09 | 6.2 | 16.6 | 2.5 | 1.7 | 6.7 | 9.6 | — | — | — | — | — | — | |
| Meta R-CNNshot=10, Backbone=ResNet-502019.09 | 5.6 | 14.2 | 3 | 2 | 6.6 | 8.8 | — | — | — | — | — | — | |
| YOLO-Few-shotShot=10, Pipeline=FRCN pipeline, Backbone=ResNet-502019.09 | 5.6 | 12.3 | 4.6 | 90 | 350 | 1,050 | 10.1 | 14.3 | 14.4 | 150 | 840 | 2,820 | |
| Feature Reweightingdataset=COCO, fine-tune=MS COCO2019.08 | 5.6 | 12.3 | 4.6 | — | — | — | — | — | — | — | — | — | |
| YOLO-FSShot=102020.07 | 5.6 | 12.3 | 4.6 | 90 | 350 | 10.5 | 10.1 | 14.3 | 14.4 | 150 | 8.4 | 28.2 | |
| MRCN+ft-fullshot=20, Backbone=ResNet-502019.09 | 4.5 | 9.8 | 3.4 | 2 | 4.6 | 6.2 | — | — | — | — | — | — | |
| Meta R-CNNshot=5, Backbone=ResNet-502019.09 | 3.5 | 9.9 | 1.2 | 1.2 | 3.9 | 5.8 | — | — | — | — | — | — | |
| MRCN+ft-fullshot=10, Backbone=ResNet-502019.09 | 2.5 | 5.7 | 1.9 | 2 | 2.7 | 3.9 | — | — | — | — | — | — | |
| FRCN+ftShot=30, Pipeline=FRCN pipeline, Backbone=ResNet-502019.09 | 1.5 | 4.8 | 0.5 | 30 | 180 | 200 | 7 | 10.1 | 10.1 | 580 | 830 | 1,350 | |
| MRCN+ft-fullshot=5, Backbone=ResNet-502019.09 | 1.3 | 3 | 1.1 | 0.3 | 1.1 | 2.4 | — | — | — | — | — | — | |
| FRCN+ftShot=10, Pipeline=FRCN pipeline, Backbone=ResNet-502019.09 | 1.3 | 4.2 | 0.4 | 40 | 90 | 210 | 5.5 | 8 | 8 | 240 | 640 | 1,300 |