Hand Instance Segmentation on HaDR real camera (test)
52.3AP (IoU 0.5:0.95)Mask R-CNN
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Mask R-CNNBackbone=ResNet50, Training Modality=RGB, Minimum confidence score threshold=0.12023.04 | 52.3 | 82.1 | 45.9 | 61.2 | 57.6 | 35.7 | |
| SOLOv2Backbone=ResNet101, Training Modality=RGB, Minimum confidence score threshold=0.12023.04 | 47.4 | 74 | 39.1 | 58.3 | 55 | 33.6 | |
| SOLOv2Backbone=ResNet50, Training Modality=RGB, Minimum confidence score threshold=0.12023.04 | 47.3 | 71.7 | 39.9 | 57.8 | 55.7 | 33 | |
| Mask R-CNNBackbone=ResNet50, Training Modality=RGB-D, Minimum confidence score threshold=0.12023.04 | 44.9 | 80 | 41.2 | 50.5 | 47.9 | 32.5 | |
| Mask R-CNNBackbone=ResNet101, Training Modality=RGB-D, Minimum confidence score threshold=0.12023.04 | 44.5 | 77.9 | 42.3 | 50.4 | 46.8 | 32.3 | |
| Mask R-CNNBackbone=ResNet101, Training Modality=RGB, Minimum confidence score threshold=0.12023.04 | 44.3 | 70.4 | 38.8 | 56.7 | 49.7 | 32.1 | |
| SOLOv2Backbone=ResNet101, Training Modality=RGB-D, Minimum confidence score threshold=0.12023.04 | 41 | 70.9 | 34.8 | 45.7 | 45.8 | 30.6 | |
| SOLOv2Backbone=ResNet50, Training Modality=RGB-D, Minimum confidence score threshold=0.12023.04 | 40.8 | 71.8 | 32.6 | 46.5 | 46 | 30.6 | |
| Mask R-CNNBackbone=ResNet50, Training Modality=Depth, Minimum confidence score threshold=0.12023.04 | 36.4 | 71.2 | 29.1 | 42.4 | 41.5 | 27.3 | |
| Mask R-CNNBackbone=ResNet101, Training Modality=Depth, Minimum confidence score threshold=0.12023.04 | 35.7 | 68.6 | 30.8 | 41.5 | 39.4 | 27.4 | |
| SOLOv2Backbone=ResNet101, Training Modality=Depth, Minimum confidence score threshold=0.12023.04 | 35.6 | 68.1 | 29.3 | 39.2 | 41.5 | 27.4 | |
| SOLOv2Backbone=ResNet50, Training Modality=Depth, Minimum confidence score threshold=0.12023.04 | 33.8 | 64.4 | 26.3 | 36.9 | 40.5 | 26.5 | |
| SOLOv2Backbone=ResNet50, Training Modality=COCO, Minimum confidence score threshold=0.12023.04 | 32.9 | 60.7 | 39.4 | 49 | 27.4 | 28.1 | |
| SOLOv2Backbone=ResNet101, Training Modality=COCO, Minimum confidence score threshold=0.12023.04 | 25.2 | 57.4 | 27.8 | 34.9 | 24.4 | 22.5 | |
| Mask R-CNNBackbone=ResNet50, Training Modality=COCO, Minimum confidence score threshold=0.12023.04 | 12.3 | 29.2 | 21.8 | 32.2 | 8.2 | 15.7 | |
| Mask R-CNNBackbone=ResNet101, Training Modality=COCO, Minimum confidence score threshold=0.12023.04 | 10.1 | 23.4 | 22.1 | 28.4 | 8.2 | 14.8 |