Object Detection on MS-COCO 2% labeled protocol
27.7mAPVirtual Category Learning (with tricks)
Evaluation Results
| Method | Links | |
|---|---|---|
| Virtual Category Learning (with tricks)Configuration=Training tricks in Soft Teacher2022.07 | 27.7 | |
| Soft TeacherOfficial code=true2022.07 | 25.93 | |
| Virtual Category LearningConfiguration=Same settings as Unbiased Teacher2022.07 | 25.84 | |
| Unbiased Teacher2022.07 | 24.3 | |
| Region Uncertainty QuantificationStrong data augmentation=true2021.11 | 24 | |
| Combating Noise2022.07 | 24 | |
| Instant-Teaching*Backbone=R50-FPN, co-rectify scheme=true2021.03 | 22.45 | |
| Instant-TeachingStrong data augmentation=true2021.11 | 22.45 | |
| Instant Teaching2022.07 | 22.45 | |
| Interactive2022.07 | 22.43 | |
| Unbiased TeacherStrong data augmentation=true2021.11 | 21.98 | |
| Humble TeacherStrong data augmentation=true2021.11 | 21.72 | |
| Humble Teacher2022.07 | 21.72 | |
| Instant-TeachingBackbone=R50-FPN, co-rectify scheme=false2021.03 | 20.7 | |
| STACBackbone=ResNet-50 with FPN, Augmentation=C+{B,G}+Cutout, Training Data=Semi-supervised2020.05 | 18.25 | |
| STACBackbone=R50-FPN2021.03 | 18.25 | |
| STACStrong data augmentation=true2021.11 | 18.25 | |
| STAC2022.07 | 18.25 | |
| Supervised†Backbone=ResNet-50 with FPN, Augmentation=C+{B,G}+Cutout, Training Data=Labeled data only2020.05 | 14.28 | |
| CSDStrong data augmentation=false2021.11 | 13.93 | |
| CSD2022.07 | 13.93 | |
| CSD†Backbone=R50-FPN2021.03 | 13.6 | |
| SupervisedBackbone=ResNet-50 with FPN, Augmentation=Standard, Training Data=Labeled data only2020.05 | 12.7 | |
| SupervisedBackbone=R50-FPN2021.03 | 12.7 | |
| Supervised2022.07 | 12.7 |