Object Detection on COCO (AP Box, Δ Full)
53.4AP BoxMONA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MONABackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=4.16 M, Trainable backbone parameters %=4.67%, Extra Structure=true2024.08 | 53.4 | 1 | |
| FULLBackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=89.14 M, Trainable backbone parameters %=100.00%, Extra Structure=false2024.08 | 52.4 | — | |
| ADAPTERBackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=3.19 M, Trainable backbone parameters %=3.58%, Extra Structure=true2024.08 | 52.1 | -0.3 | |
| LORANDBackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=4.68 M, Trainable backbone parameters %=5.23%, Extra Structure=true2024.08 | 51.9 | -0.5 | |
| ADAPTFORMERBackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=1.60 M, Trainable backbone parameters %=1.79%, Extra Structure=true2024.08 | 51.7 | -0.7 | |
| PARTIAL-1Backbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=12.95 M, Trainable backbone parameters %=14.53%, Extra Structure=false2024.08 | 50.6 | -1.8 | |
| LORABackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=3.06 M, Trainable backbone parameters %=3.43%, Extra Structure=true2024.08 | 50.4 | -2 | |
| BITFITBackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=0.21 M, Trainable backbone parameters %=0.23%, Extra Structure=false2024.08 | 50.1 | -2.3 | |
| NORMTUNINGBackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=0.06 M, Trainable backbone parameters %=0.07%, Extra Structure=false2024.08 | 50.1 | -2.3 | |
| FIXEDBackbone=Swin-B (89M), Framework=Cascade Mask R-CNN, Trained Params=0.00 M, Trainable backbone parameters %=0.00%, Extra Structure=false2024.08 | 48 | -4.4 |