Category-agnostic instance segmentation on WISDOM-Real Low-Res Primesense camera (test)
38.5APFT Mask R-CNN (Color)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FT Mask R-CNN (Color)Backbone=ResNet-101 / FPN, Input modality=Color, Pre-trained=COCO RGB, Fine-tuning protocol=100 images from WISDOM-Real High-Res (Color)2018.09 | 38.5 | 61.3 | |
| SD Mask R-CNNBackbone=ResNet-101 / FPN, Input modality=Depth, Training data source=WISDOM-Sim2018.09 | 35.6 | 46.5 | |
| FT Mask R-CNN (Depth)Backbone=ResNet-101 / FPN, Input modality=Depth, Pre-trained=COCO RGB, Fine-tuning protocol=100 images from WISDOM-Real High-Res (Depth)2018.09 | 33.1 | 54.6 | |
| Euclidean ClusteringBackbone=Point Cloud Library (PCL), Input modality=Depth (Point Cloud), Training data source=N/A (Heuristic)2018.09 | 18.3 | 31.7 | |
| Region GrowingBackbone=Point Cloud Library (PCL), Input modality=Depth (Point Cloud), Training data source=N/A (Heuristic)2018.09 | 18 | 34.6 |