Unsupervised Panoptic Segmentation
Unsupervised Panoptic Segmentation aims to partition an image into semantically meaningful regions and distinct object instances without training using any manually annotated images. ( Image credit: CUPS, Hahn & Reich et al., CVPR 2025 )
Papers
Showing 1–4 of 4 papers
All datasetsCityscapesBDD100K valKITTIMUSES: MUlti-SEnsor Semantic perception datasetWaymo Open DatasetCOCO val2017
Benchmark Results
| # | Model | Metric | Claimed | Verified | Status |
|---|---|---|---|---|---|
| 1 | CUPS (54 pseudo-classes) | PQ | 30.6 | — | Unverified |
| 2 | CUPS (40 pseudo-classes) | PQ | 30.3 | — | Unverified |
| 3 | CUPS (27 pseudo-classes) | PQ | 27.8 | — | Unverified |
| 4 | U2Seg (827 pseudo-classes) | PQ | 18.4 | — | Unverified |
| 5 | DepthG + CutLER | PQ | 16.1 | — | Unverified |