Object Detection on MS COCO novel classes 2017 (val)
22.73APOurs (Decoupling Classifier)
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Ours (Decoupling Classifier)K-shot=302025.05 | 22.73 | 40.24 | — | — | — | — | — | — | — | — | — | — | |
| Mask-DeFRCN†K-shot=302025.05 | 22.63 | 38.95 | — | — | — | — | — | — | — | — | — | — | |
| Meta-DETRShot=30, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 22.2 | 35 | 22.8 | — | — | — | — | — | — | — | — | — | |
| Ours (Decoupling Classifier)K-shot=102025.05 | 19.33 | 34.78 | — | — | — | — | — | — | — | — | — | — | |
| Meta-DETRShot=10, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 19 | 30.5 | 19.7 | — | — | — | — | — | — | — | — | — | |
| Mask-DeFRCN†K-shot=102025.05 | 18.72 | 32.8 | — | — | — | — | — | — | — | — | — | — | |
| DCNetShot=30, Multi-scale features=true, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 18.6 | 32.6 | 17.5 | — | — | — | — | — | — | — | — | — | |
| CMEShot=30, Multi-scale features=true, Backbone=ResNet-1012021.03 | 16.9 | 28 | 17.8 | — | — | — | — | — | — | — | — | — | |
| Ours (Decoupling Classifier)K-shot=52025.05 | 16.39 | 29.96 | — | — | — | — | — | — | — | — | — | — | |
| Deformable-DETR-ft-fullShot=30, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 16.3 | 27.2 | 16.7 | — | — | — | — | — | — | — | — | — | |
| Meta-DETRShot=5, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 15.4 | 25 | 15.8 | — | — | — | — | — | — | — | — | — | |
| Mask-DeFRCN†K-shot=52025.05 | 15.39 | 27.29 | — | — | — | — | — | — | — | — | — | — | |
| FSCEShot=30, Multi-scale features=true, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 15.3 | — | 14.2 | — | — | — | — | — | — | — | — | — | |
| CMEShot=10, Multi-scale features=true, Backbone=ResNet-1012021.03 | 15.1 | 24.6 | 16.4 | — | — | — | — | — | — | — | — | — | |
| FsDetViewShot=30, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 14.7 | 30.6 | 12.2 | — | — | — | — | — | — | — | — | — | |
| SRR-FSDShot=30, Multi-scale features=true, Backbone=ResNet-1012021.03 | 14.7 | 29.2 | 13.5 | — | — | — | — | — | — | — | — | — | |
| DAnA-FasterRCNN# Way=1-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 14.4 | 30.4 | 13 | — | — | — | — | — | — | — | — | — | |
| DAnA-FasterRCNN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=5shot, # parameters=1.42 x 10^8, FPS=242021.02 | 14.4 | 30.4 | 13 | — | — | — | — | — | — | — | — | — | |
| MPSRShot=30, Multi-scale features=true, Backbone=ResNet-1012021.03 | 14.1 | 25.4 | 14.2 | — | — | — | — | — | — | — | — | — | |
| Ours (Decoupling Classifier)K-shot=32025.05 | 14.04 | 25.74 | — | — | — | — | — | — | — | — | — | — | |
| DAnA-FasterRCNN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=3shot, # parameters=1.42 x 10^8, FPS=242021.02 | 14 | 28.9 | 12.3 | — | — | — | — | — | — | — | — | — | |
| DAnA-FasterRCNN# Way=3-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 13.7 | 28.2 | 12.4 | — | — | — | — | — | — | — | — | — | |
| Meta-DETRShot=3, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 13.5 | 21.7 | 14 | — | — | — | — | — | — | — | — | — | |
| IR-FSODBackbone=Res-50, Tuning time=IR, Shot setting=10-shot2021.10 | 13.1 | 24.5 | 12.3 | 5.9 | 16.2 | 22 | 19.1 | 33.5 | 35.6 | 11.4 | 39.4 | 54.9 | |
| Mask-DeFRCN†K-shot=32025.05 | 13.07 | 23.28 | — | — | — | — | — | — | — | — | — | — | |
| DCNetShot=10, Multi-scale features=true, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 12.8 | 23.4 | 11.2 | — | — | — | — | — | — | — | — | — | |
| DCNetBackbone=Res-101, Tuning time=40+ min, Shot setting=10-shot2021.10 | 12.8 | 23.4 | 11.2 | 4.3 | 13.8 | 21 | 18.1 | 26.7 | 25.6 | 7.9 | 24.5 | 36.7 | |
| DAnA-FasterRCNN# Way=5-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 12.6 | 25.9 | 11.3 | — | — | — | — | — | — | — | — | — | |
| FsDetViewShot=10, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 12.5 | 27.3 | 9.8 | — | — | — | — | — | — | — | — | — | |
| FSIWBackbone=Res-50, Tuning time=5 h, Shot setting=10-shot2021.10 | 12.5 | 27.3 | 9.8 | 2.5 | 13.8 | 19.9 | 20 | 25.5 | 25.7 | 7.5 | 27.6 | 38.9 | |
| Meta R-CNNShot=30, Backbone=ResNet-1012021.03 | 12.4 | 25.3 | 10.8 | — | — | — | — | — | — | — | — | — | |
| TFA w/ cosShot=30, Multi-scale features=true, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 12.1 | 22 | 12 | — | — | — | — | — | — | — | — | — | |
| FSODShot=10, Backbone=ResNet-1012021.03 | 12 | 22.4 | 11.8 | — | — | — | — | — | — | — | — | — | |
| A-RPN+FTBackbone=Res-50, Tuning time=15 min, Shot setting=10-shot2021.10 | 12 | 22.4 | 11.8 | 2.9 | 12.2 | 20.7 | 18.8 | 26.4 | 26.4 | 3.6 | 23.6 | 45.6 | |
| IR-FSODBackbone=Res-50, Shots=10-shot2021.10 | 12 | 23.5 | 10.8 | — | — | — | — | — | — | — | — | — | |
| FSCEBackbone=FPN-101, Tuning time=3 h, Shot setting=10-shot2021.10 | 11.9 | 22.3 | 11.6 | 2.9 | 11.1 | 17.6 | 17 | 26.6 | 26.5 | 6.7 | 26.3 | 42.3 | |
| DAnA-FasterRCNN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=1shot, # parameters=1.42 x 10^8, FPS=242021.02 | 11.9 | 25.6 | 10.4 | — | — | — | — | — | — | — | — | — | |
| Ours (Decoupling Classifier)K-shot=22025.05 | 11.9 | 22.39 | — | — | — | — | — | — | — | — | — | — | |
| Deformable-DETR-ft-fullShot=10, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 11.7 | 19.6 | 12.1 | — | — | — | — | — | — | — | — | — | |
| SRR-FSDShot=10, Multi-scale features=true, Backbone=ResNet-1012021.03 | 11.3 | 23 | 9.8 | — | — | — | — | — | — | — | — | — | |
| Meta DetShot=30, Backbone=ResNet-1012021.03 | 11.3 | 21.7 | 8.1 | — | — | — | — | — | — | — | — | — | |
| FSCEBackbone=FPN-101, Shots=10-shot, Status=Re-implemented2021.10 | 11.2 | 22.3 | 9.8 | — | — | — | — | — | — | — | — | — | |
| Meta R-CNN# Way=1-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 11.2 | 25.9 | 8.6 | — | — | — | — | — | — | — | — | — | |
| Meta R-CNN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=5shot, # parameters=4.76 x 10^7, FPS=282021.02 | 11.2 | 25.9 | 8.6 | — | — | — | — | — | — | — | — | — | |
| FSCEShot=10, Multi-scale features=true, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 11.1 | — | 9.8 | — | — | — | — | — | — | — | — | — | |
| Meta R-CNN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=3shot, # parameters=4.76 x 10^7, FPS=282021.02 | 11.1 | 25.3 | 8.5 | — | — | — | — | — | — | — | — | — | |
| Mask-DeFRCN†K-shot=22025.05 | 11.01 | 20.2 | — | — | — | — | — | — | — | — | — | — | |
| Meta R-CNN# Way=3-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 11 | 25.2 | 8.6 | — | — | — | — | — | — | — | — | — | |
| A-RPN+FTBackbone=Res-50, Shots=10-shot, Status=Re-implemented2021.10 | 10.9 | 20.5 | 9.4 | — | — | — | — | — | — | — | — | — | |
| FGN+# Way=1-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 10.9 | 24 | 9 | — | — | — | — | — | — | — | — | — | |
| FGN+# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=5shot, # parameters=1.48 x 10^8, FPS=232021.02 | 10.9 | 24 | 8.8 | — | — | — | — | — | — | — | — | — | |
| FGN+# Way=3-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 10.8 | 23.4 | 9.1 | — | — | — | — | — | — | — | — | — | |
| FsDetViewShot=5, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 10.7 | 24.5 | 6.7 | — | — | — | — | — | — | — | — | — | |
| FSIWBackbone=Res-50, Shots=10-shot2021.10 | 10.7 | 25.6 | 6.5 | — | — | — | — | — | — | — | — | — | |
| Attention RPN# Way=1-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 10.6 | 24.4 | 8.3 | — | — | — | — | — | — | — | — | — | |
| Attention RPN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=5shot, # parameters=1.03 x 10^8, FPS=212021.02 | 10.6 | 24.4 | 8.3 | — | — | — | — | — | — | — | — | — | |
| IR-FSODBackbone=Res-50, Shots=5-shot2021.10 | 10.5 | 20.8 | 9.1 | — | — | — | — | — | — | — | — | — | |
| FGN+# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=3shot, # parameters=1.48 x 10^8, FPS=232021.02 | 10.5 | 22.5 | 8.6 | — | — | — | — | — | — | — | — | — | |
| Meta R-CNN# Way=5-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 10.2 | 23.4 | 8 | — | — | — | — | — | — | — | — | — | |
| Attention RPN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=3shot, # parameters=1.03 x 10^8, FPS=212021.02 | 10.1 | 23 | 8.2 | — | — | — | — | — | — | — | — | — | |
| MPSRShot=10, Multi-scale features=true, Backbone=ResNet-1012021.03 | 9.8 | 17.9 | 9.7 | — | — | — | — | — | — | — | — | — | |
| MPSRBackbone=FPN-101, Tuning time=40+ min, Shot setting=10-shot2021.10 | 9.8 | 17.9 | 9.7 | 3.3 | 9.2 | 16.1 | 15.7 | 21.2 | 21.2 | 4.6 | 19.6 | 34.3 | |
| TFABackbone=FPN-101, Tuning time=16 h, Shot setting=10-shot2021.10 | 9.8 | 19.7 | 8.9 | 2.8 | 9.2 | 16.1 | 14.5 | 18.6 | 18.6 | 5.3 | 14.8 | 33.1 | |
| Attention RPN# Way=3-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 9.8 | 22.8 | 7.9 | — | — | — | — | — | — | — | — | — | |
| FGN+# Way=5-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 9.6 | 21.2 | 8.1 | — | — | — | — | — | — | — | — | — | |
| TFA w/ cosShot=10, Multi-scale features=true, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 9.1 | 17.1 | 8.8 | — | — | — | — | — | — | — | — | — | |
| Meta-YOLOShot=30, Backbone=ResNet-1012021.03 | 9.1 | 19 | 7.6 | — | — | — | — | — | — | — | — | — | |
| TFABackbone=FPN-101, Shots=10-shot2021.10 | 9.1 | 17.1 | 8.8 | — | — | — | — | — | — | — | — | — | |
| Attention RPN# Way=5-way, # Given Supports=5-shot, Evaluation Protocol=Zero-shot2021.02 | 9 | 20.8 | 7.3 | — | — | — | — | — | — | — | — | — | |
| IR-FSODBackbone=Res-50, Shots=3-shot2021.10 | 8.9 | 17.6 | 7.9 | — | — | — | — | — | — | — | — | — | |
| Meta R-CNNShot=10, Backbone=ResNet-1012021.03 | 8.7 | 19.1 | 6.6 | — | — | — | — | — | — | — | — | — | |
| Meta R-CNNBackbone=Res-50, Tuning time=5+ h, Shot setting=10-shot2021.10 | 8.7 | 19.1 | 6.6 | 2.3 | 7.7 | 14 | 12.6 | 17.8 | 17.9 | 7.8 | 15.6 | 27.2 | |
| Meta R-CNN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=1shot, # parameters=4.76 x 10^7, FPS=282021.02 | 8.7 | 19.9 | 6.8 | — | — | — | — | — | — | — | — | — | |
| Attention RPN# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=1shot, # parameters=1.03 x 10^8, FPS=212021.02 | 8.7 | 19.8 | 7 | — | — | — | — | — | — | — | — | — | |
| MTFAK-shot=102025.05 | 8.52 | 15.53 | — | — | — | — | — | — | — | — | — | — | |
| FSIWBackbone=Res-50, Shots=5-shot2021.10 | 8.1 | 20.1 | 4.4 | — | — | — | — | — | — | — | — | — | |
| A-RPN+FTBackbone=Res-50, Shots=5-shot, Status=Re-implemented2021.10 | 8.1 | 15.5 | 7.4 | — | — | — | — | — | — | — | — | — | |
| Ours (Decoupling Classifier)K-shot=12025.05 | 8.09 | 15.85 | — | — | — | — | — | — | — | — | — | — | |
| FGN+# Way=1, Evaluation Protocol=Zero-shot, # Given Supports=1shot, # parameters=1.48 x 10^8, FPS=232021.02 | 8 | 17.3 | 6.9 | — | — | — | — | — | — | — | — | — | |
| IR-FSODBackbone=Res-50, Shots=2-shot2021.10 | 7.7 | 15.7 | 6.6 | — | — | — | — | — | — | — | — | — | |
| FSCEBackbone=FPN-101, Shots=5-shot, Status=Re-implemented2021.10 | 7.6 | 15.6 | 6.5 | — | — | — | — | — | — | — | — | — | |
| Mask-DeFRCN†K-shot=12025.05 | 7.54 | 14.46 | — | — | — | — | — | — | — | — | — | — | |
| Meta-DETRShot=1, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 7.5 | 12.5 | 7.7 | — | — | — | — | — | — | — | — | — | |
| Deformable-DETR-ft-fullShot=5, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 7.4 | 12.3 | 7.7 | — | — | — | — | — | — | — | — | — | |
| FRCN-ft-fullShot=30, Multi-scale features=true, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 7.4 | 13.1 | 7.4 | — | — | — | — | — | — | — | — | — | |
| A-RPNBackbone=Res-50, Tuning time=IR, Shot setting=10-shot2021.10 | 7.3 | 13.2 | 7.1 | 4.4 | 8.7 | 10.7 | 17.5 | 32.3 | 33.2 | 10 | 34.7 | 50.4 | |
| iMTFAK-shot=102025.05 | 7.14 | 12.91 | — | — | — | — | — | — | — | — | — | — | |
| Meta DetShot=10, Backbone=ResNet-1012021.03 | 7.1 | 14.6 | 6.1 | — | — | — | — | — | — | — | — | — | |
| TFA w/ cosShot=5, Multi-scale features=true, Averaged on multiple runs=true, Backbone=ResNet-1012021.03 | 7 | 13.3 | 6.5 | — | — | — | — | — | — | — | — | — | |
| TFABackbone=FPN-101, Shots=5-shot2021.10 | 7 | 13.3 | 6.5 | — | — | — | — | — | — | — | — | — | |
| TFA w/ cos + HallucShot=3, Multi-scale features=true, Backbone=ResNet-1012021.03 | 6.9 | 12.6 | 7 | — | — | — | — | — | — | — | — | — | |
| A-RPNBackbone=Res-50, Shots=10-shot, Status=Re-implemented2021.10 | 6.7 | 12.5 | 5.8 | — | — | — | — | — | — | — | — | — | |
| FSIWBackbone=Res-50, Shots=3-shot2021.10 | 6.7 | 18.6 | 2.9 | — | — | — | — | — | — | — | — | — | |
| MTFAK-shot=52025.05 | 6.61 | 12.32 | — | — | — | — | — | — | — | — | — | — | |
| A-RPNBackbone=Res-50, Shots=5-shot, Status=Re-implemented2021.10 | 6.3 | 11.9 | 5.9 | — | — | — | — | — | — | — | — | — | |
| iMTFAK-shot=52025.05 | 6.22 | 11.28 | — | — | — | — | — | — | — | — | — | — | |
| A-RPN+FTBackbone=Res-50, Shots=3-shot, Status=Re-implemented2021.10 | 5.9 | 11.6 | 5.7 | — | — | — | — | — | — | — | — | — | |
| FSCEBackbone=FPN-101, Shots=3-shot, Status=Re-implemented2021.10 | 5.7 | 12 | 4.7 | — | — | — | — | — | — | — | — | — | |
| Meta-YOLOShot=10, Backbone=ResNet-1012021.03 | 5.6 | 12.3 | 4.6 | — | — | — | — | — | — | — | — | — | |
| A-RPNBackbone=Res-50, Shots=3-shot, Status=Re-implemented2021.10 | 5.6 | 10.7 | 5.2 | — | — | — | — | — | — | — | — | — |