Unsupervised Semantic Segmentation
Models that learn to segment each image (i.e. assign a class to every pixel) without seeing the ground truth labels.
( Image credit: SegSort: Segmentation by Discriminative Sorting of Segments )
Papers
Showing 1–10 of 95 papers
All datasetsCOCO-Stuff-27Cityscapes testPASCAL VOC 2012 valPotsdam-3COCO-Stuff-3ImageNet-S-50COCO-Stuff-171COCO-Stuff-81SUIMCOCO-Stuff-15ACDC (Adverse Conditions Dataset with Correspondences)Cityscapes val
Benchmark Results
| # | Model | Metric | Claimed | Verified | Status |
|---|---|---|---|---|---|
| 1 | SGSeg | Clustering [Accuracy] | 55.7 | — | Unverified |
| 2 | DynaSeg - FSF (ResNet-18 FPN) | Clustering [mIoU] | 54.1 | — | Unverified |
| 3 | SAN | Clustering [Accuracy] | 52 | — | Unverified |
| 4 | DiffCut | Clustering [mIoU] | 49.1 | — | Unverified |
| 5 | PiCIE | Clustering [Accuracy] | 48.1 | — | Unverified |
| 6 | HCL (ViT-S/8) | Linear Classifier [mIoU] | 47.4 | — | Unverified |
| 7 | HCL (ViT-S/16) | Linear Classifier [mIoU] | 45.8 | — | Unverified |
| 8 | CAUSE (DINOv2, ViT-B/14) | Clustering [mIoU] | 45.3 | — | Unverified |
| 9 | Ours (SlotCon) | Clustering [Accuracy] | 42.36 | — | Unverified |
| 10 | CAUSE (ViT-B/8) | Clustering [mIoU] | 41.9 | — | Unverified |