Boyu Yang, Chang Liu, Bohao Li, Jianbin Jiao, Qixiang Ye
Few-shot segmentation is challenging because objects within the support and query images could significantly differ in appearance and pose. Using a single prototype acquired directly from the support image to segment the query image causes semantic ambiguity. In this paper, we propose prototype mixture models (PMMs), which correlate diverse image regions with multiple prototypes to enforce the prototype-based semantic representation. Estimated by an Expectation-Maximization algorithm, PMMs incorporate rich channel-wised and spatial semantics from limited support images. Utilized as representations as well as classifiers, PMMs fully leverage the semantics to activate objects in the query image while depressing background regions in a duplex manner. Extensive experiments on Pascal VOC and MS-COCO datasets show that PMMs significantly improve upon state-of-the-arts. Particularly, PMMs improve 5-shot segmentation performance on MS-COCO by up to 5.82\% with only a moderate cost for model size and inference speed.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Few-Shot Learning | COCO-20i (5-shot) | Mean IoU | 35.5 | RPMM (ResNet-50) |
| Few-Shot Learning | COCO-20i -> Pascal VOC (1-shot) | Mean IoU | 49.6 | RPMM |
| Few-Shot Learning | PASCAL-5i (10-Shot) | Mean IoU | 57.6 | RPMM |
| Few-Shot Learning | COCO-20i (10-shot) | Mean IoU | 33.1 | RPMM |
| Few-Shot Learning | PASCAL-5i (1-Shot) | Mean IoU | 56.3 | RPMM (ResNet-50) |
| Few-Shot Learning | COCO-20i (1-shot) | Mean IoU | 30.6 | RPMM (ResNet-50) |
| Few-Shot Learning | PASCAL-5i (5-Shot) | Mean IoU | 57.3 | RPMM (ResNet-50) |
| Few-Shot Learning | COCO-20i -> Pascal VOC (5-shot) | Mean IoU | 53.8 | RPMM |
| Few-Shot Semantic Segmentation | COCO-20i (5-shot) | Mean IoU | 35.5 | RPMM (ResNet-50) |
| Few-Shot Semantic Segmentation | COCO-20i -> Pascal VOC (1-shot) | Mean IoU | 49.6 | RPMM |
| Few-Shot Semantic Segmentation | PASCAL-5i (10-Shot) | Mean IoU | 57.6 | RPMM |
| Few-Shot Semantic Segmentation | COCO-20i (10-shot) | Mean IoU | 33.1 | RPMM |
| Few-Shot Semantic Segmentation | PASCAL-5i (1-Shot) | Mean IoU | 56.3 | RPMM (ResNet-50) |
| Few-Shot Semantic Segmentation | COCO-20i (1-shot) | Mean IoU | 30.6 | RPMM (ResNet-50) |
| Few-Shot Semantic Segmentation | PASCAL-5i (5-Shot) | Mean IoU | 57.3 | RPMM (ResNet-50) |
| Few-Shot Semantic Segmentation | COCO-20i -> Pascal VOC (5-shot) | Mean IoU | 53.8 | RPMM |
| Meta-Learning | COCO-20i (5-shot) | Mean IoU | 35.5 | RPMM (ResNet-50) |
| Meta-Learning | COCO-20i -> Pascal VOC (1-shot) | Mean IoU | 49.6 | RPMM |
| Meta-Learning | PASCAL-5i (10-Shot) | Mean IoU | 57.6 | RPMM |
| Meta-Learning | COCO-20i (10-shot) | Mean IoU | 33.1 | RPMM |
| Meta-Learning | PASCAL-5i (1-Shot) | Mean IoU | 56.3 | RPMM (ResNet-50) |
| Meta-Learning | COCO-20i (1-shot) | Mean IoU | 30.6 | RPMM (ResNet-50) |
| Meta-Learning | PASCAL-5i (5-Shot) | Mean IoU | 57.3 | RPMM (ResNet-50) |
| Meta-Learning | COCO-20i -> Pascal VOC (5-shot) | Mean IoU | 53.8 | RPMM |