Panoptic Segmentation on COCO 2017
0.514PQFull-tuning
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Full-tuningBackbone=Swin-L, Number of Parameters (M)=195.02026.02 | 0.514 | 0.815 | 0.619 | |
| Full-tuningBackbone=ConvNeXt-L, Number of Parameters (M)=196.22026.02 | 0.51 | 0.809 | 0.614 | |
| AdaRouteBackbone=ConvNeXt-L, Number of Parameters (M)=9.22026.02 | 0.504 | 0.812 | 0.607 | |
| Full-tuningBackbone=Swin-B, Number of Parameters (M)=86.82026.02 | 0.503 | 0.813 | 0.606 | |
| AdaRouteBackbone=Swin-L, Number of Parameters (M)=7.32026.02 | 0.502 | 0.813 | 0.605 | |
| Full-tuningBackbone=ConvNeXt-B, Number of Parameters (M)=87.62026.02 | 0.502 | 0.812 | 0.605 | |
| MonaBackbone=Swin-L, Number of Parameters (M)=7.52026.02 | 0.497 | 0.807 | 0.602 | |
| MonaBackbone=ConvNeXt-L, Number of Parameters (M)=9.12026.02 | 0.495 | 0.809 | 0.597 | |
| CoLoRABackbone=ConvNeXt-L, Number of Parameters (M)=9.42026.02 | 0.493 | 0.805 | 0.596 | |
| RepAdapterBackbone=Swin-L, Number of Parameters (M)=8.72026.02 | 0.492 | 0.807 | 0.596 | |
| AdaRouteBackbone=ConvNeXt-B, Number of Parameters (M)=6.52026.02 | 0.489 | 0.81 | 0.591 | |
| AdaRouteBackbone=Swin-B, Number of Parameters (M)=5.22026.02 | 0.488 | 0.808 | 0.59 | |
| AdaptFormerBackbone=Swin-L, Number of Parameters (M)=8.12026.02 | 0.488 | 0.799 | 0.592 | |
| MonaBackbone=ConvNeXt-B, Number of Parameters (M)=6.52026.02 | 0.483 | 0.801 | 0.585 | |
| MonaBackbone=Swin-B, Number of Parameters (M)=5.22026.02 | 0.481 | 0.799 | 0.583 | |
| RepAdapterBackbone=Swin-B, Number of Parameters (M)=5.82026.02 | 0.479 | 0.799 | 0.581 | |
| LoRABackbone=ConvNeXt-L, Number of Parameters (M)=8.82026.02 | 0.478 | 0.802 | 0.578 | |
| RepAdapterBackbone=ConvNeXt-L, Number of Parameters (M)=9.12026.02 | 0.478 | 0.799 | 0.579 | |
| LoRandBackbone=Swin-L, Number of Parameters (M)=8.22026.02 | 0.477 | 0.803 | 0.58 | |
| CoLoRABackbone=ConvNeXt-B, Number of Parameters (M)=6.42026.02 | 0.476 | 0.796 | 0.579 | |
| SNELLBackbone=Swin-L, Number of Parameters (M)=8.72026.02 | 0.472 | 0.794 | 0.576 | |
| AdaptFormerBackbone=ConvNeXt-L, Number of Parameters (M)=9.62026.02 | 0.472 | 0.796 | 0.573 | |
| AdaptFormerBackbone=Swin-B, Number of Parameters (M)=5.42026.02 | 0.471 | 0.794 | 0.574 | |
| RepAdapterBackbone=ConvNeXt-B, Number of Parameters (M)=6.12026.02 | 0.471 | 0.797 | 0.572 | |
| SNELLBackbone=ConvNeXt-L, Number of Parameters (M)=9.52026.02 | 0.469 | 0.792 | 0.571 | |
| LoRandBackbone=ConvNeXt-L, Number of Parameters (M)=9.32026.02 | 0.468 | 0.794 | 0.569 | |
| VPTBackbone=Swin-L, Number of Parameters (M)=0.22026.02 | 0.466 | 0.791 | 0.57 | |
| LoRABackbone=Swin-L, Number of Parameters (M)=8.12026.02 | 0.466 | 0.794 | 0.571 | |
| AdaptFormerBackbone=ConvNeXt-B, Number of Parameters (M)=6.42026.02 | 0.466 | 0.788 | 0.567 | |
| LoRandBackbone=Swin-B, Number of Parameters (M)=5.92026.02 | 0.464 | 0.793 | 0.567 | |
| SNELLBackbone=Swin-B, Number of Parameters (M)=5.82026.02 | 0.46 | 0.792 | 0.564 | |
| LoRABackbone=ConvNeXt-B, Number of Parameters (M)=5.92026.02 | 0.46 | 0.789 | 0.56 | |
| LoRandBackbone=ConvNeXt-B, Number of Parameters (M)=6.72026.02 | 0.459 | 0.786 | 0.56 | |
| SNELLBackbone=ConvNeXt-B, Number of Parameters (M)=6.42026.02 | 0.455 | 0.789 | 0.555 | |
| LoRABackbone=Swin-B, Number of Parameters (M)=5.42026.02 | 0.453 | 0.785 | 0.557 | |
| VPTBackbone=Swin-B, Number of Parameters (M)=0.12026.02 | 0.451 | 0.786 | 0.554 |