Research map / Continual learning / Class-incremental
SSUL: Semantic Segmentation with Unknown Label for Exemplar-based Class-Incremental Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SSUL instead of PLOP | Class-incremental semantic segmentation · VOC 10-1, 5-1, 2-1, and 2-2 | all-class mIoU: 30.45 -> 59.25; 6.46 -> 48.65; 4.47 -> 38.32; 13.66 -> 45.31 | Helped |
| Add exemplar memory (ablation) | SSUL · VOC 15-1 | new-class mIoU: 36.59 -> 49.01 | Helped |
| Freeze the model during incremental learning (ablation) | SSUL · VOC 15-1 | mIoU | Helped |
| Add unknown-class modeling (ablation) | SSUL · VOC 15-1 | new-class mIoU | Helped |
| Add head weight transfer instead of random initialization (ablation) | SSUL · VOC 15-1 | all-class mIoU: 61.89 -> 67.61 | Helped |
| Increase random-initialization training from ×1 to ×2 iterations (ablation) | SSUL head · VOC 15-1 | all-class mIoU: 61.89 -> 62.45 | Helped |
| Increase random-initialization training from ×2 to ×4 iterations (ablation) | SSUL head · VOC 15-1 | all-class mIoU: 62.45 -> 61.14 | Hurt |
| Use saliency maps to select pseudo-label regions (ablation) | PLOP · VOC 15-1 | all-class mIoU: 54.64% → 57.01% | Helped |
| Add unknown-class modeling and saliency without correct weight transfer (ablation) | SSUL · VOC 15-1 | all-class mIoU: 61.12 -> 61.89 | No clear effect |
| Set pseudo-label confidence threshold τ to 0.7 rather than 0.0 (ablation) | SSUL · VOC 15-1 | all-class mIoU: 67.44 -> 67.81 | Helped |
| Set pseudo-label confidence threshold τ to 1.0 rather than 0.9 (ablation) | SSUL · VOC 15-1 | all-class mIoU: 67.61 -> 66.74 | Hurt |
Closest papers by meaning
- Incrementer: Transformer for Class-Incremental Semantic Segmentation With Knowledge Distillation Focusing on Old Class (CVPR 2023 · similarity 0.92)
- From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.88)
- CoMBO: Conflict Mitigation via Branched Optimization for Class Incremental Segmentation (CVPR 2025 · similarity 0.88)
- Adaptive Prototype Replay for Class Incremental Semantic Segmentation (AAAI 2025 · similarity 0.87)
- Deconfound Semantic Shift and Incompleteness in Incremental Few-shot Semantic Segmentation (AAAI 2025 · similarity 0.87)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.