FARCLUSS: Fuzzy Adaptive Rebalancing and Contrastive Uncertainty Learning for Semi-Supervised Semantic Segmentation
About
Semi-supervised semantic segmentation (SSSS) faces persistent challenges in effectively leveraging unlabeled data, such as ineffective utilization of pseudo-labels, exacerbation of class imbalance biases, and neglect of prediction uncertainty. Current approaches often discard uncertain regions through strict thresholding favouring dominant classes. To address these limitations, we introduce a holistic framework that transforms uncertainty into a learning asset through four principal components: (1) fuzzy pseudo-labeling, which preserves soft class distributions from top-K predictions to enrich supervision; (2) uncertainty-aware dynamic weighting, that modulate pixel-wise contributions via entropy-based reliability scores; (3) adaptive class rebalancing, which dynamically adjust losses to counteract long-tailed class distributions; and (4) lightweight contrastive regularization, that encourage compact and discriminative feature embeddings. Extensive experiments on benchmarks demonstrate that our method outperforms current state-of-the-art approaches, achieving significant improvements in the segmentation of under-represented classes and ambiguous regions.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Semantic segmentation | Cityscapes (val) | mIoU81 | 572 | |
| Semantic segmentation | Pascal VOC 21 classes (val) | mIoU80.3 | 103 | |
| Semantic segmentation | Pascal Blended 1/4 augmented (train) | mIoU79.8 | 32 | |
| Semantic segmentation | Pascal Blended augmented (1/8 train) | mIoU79 | 32 | |
| Semantic segmentation | Pascal Blended 662 labels augmented (1/16 train) | mIoU78.5 | 31 | |
| Semantic segmentation | Pascal VOC Classic 2012 (1/8) | Unlabeled Samples per Epoch9.2 | 7 | |
| Semantic segmentation | Pascal VOC Classic 2012 (1/4) | Unlabeled Samples/Epoch7.90e+3 | 7 | |
| Semantic segmentation | Cityscapes 1/16 | Unlabeled Samples/Epoch2.7 | 7 | |
| Semantic segmentation | Cityscapes (1/8) | Unlabeled Samples per Epoch2.6 | 7 | |
| Semantic segmentation | Cityscapes 1/4 | Unlabeled Samples/Epoch2.20e+3 | 7 |