Research map / Continual learning / Class-incremental
Endpoints Weight Fusion for Class Incremental Semantic Segmentation
Jia-Wen Xiao, Chang-Bin Zhang, Jiekang Feng, Xialei Liu, Joost van de Weijer, Ming-Ming Cheng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Fuse the old and new task endpoint weights with dynamic αt | MiB on PASCAL VOC 2012 15-1 · PASCAL VOC 2012 15-1 | mIoU (%): 32.2 -> 65.5 | Improved |
| Fuse the old and new task endpoint weights with dynamic αt | PLOP on PASCAL VOC 2012 10-1 · PASCAL VOC 2012 10-1 | mIoU (%): 30.5 -> 51.9 | Improved |
| Fuse the old and new task endpoint weights with dynamic αt | MiB on ADE20K 100-5 · ADE20K 100-5 | mIoU (%): 25.9 -> 32.1 | Improved |
| Add EWF to SSUL | SSUL on PASCAL VOC 2012 15-1 · PASCAL VOC 2012 15-1 | mIoU (%): 67.6 -> 68.6 | Improved |
| Replace EWF with EMA fusion (ablation) | MiB on PASCAL VOC 2012 15-1 · PASCAL VOC 2012 15-1 | mIoU (%): 65.6 -> 37.3 | Hurt |
| Replace EWF with prediction-level model ensemble (ablation) | MiB on PASCAL VOC 2012 15-1 · PASCAL VOC 2012 15-1 | mIoU (%): 65.6 -> 37.2 | Hurt |
| Remove model fusion (ablation) | MiB on PASCAL VOC 2012 15-1 · PASCAL VOC 2012 15-1 | mIoU (%): 65.6 -> 32.2 | Hurt |
| Replace dynamic αt with fixed α=0.2 (ablation) | MiB+EWF · PASCAL VOC 2012 settings 15-1, 10-1, and 5-3 | average mIoU (%): 51.6 -> 47.7 | Hurt |
| Replace dynamic αt with fixed α=0.4 (ablation) | MiB+EWF · PASCAL VOC 2012 settings 15-1, 10-1, and 5-3 | average mIoU (%): 51.6 -> 48.9 | Hurt |
| Replace dynamic αt with fixed α=0.6 (ablation) | MiB+EWF · PASCAL VOC 2012 settings 15-1, 10-1, and 5-3 | average mIoU (%): 51.6 -> 46.3 | Hurt |
| Replace dynamic αt with fixed α=0.8 (ablation) | MiB+EWF · PASCAL VOC 2012 settings 15-1, 10-1, and 5-3 | average mIoU (%): 51.6 -> 40.2 | Hurt |
| Fuse endpoint weights with dynamic αt | MiB on PASCAL VOC 2012 19-1 · PASCAL VOC 2012 19-1 | mIoU (%): 74.7 -> 74.7 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SSUL — SSUL: Semantic Segmentation with Unknown Label for Exemplar-based Class-Incremental Learning (NeurIPS 2021) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Incrementer: Transformer for Class-Incremental Semantic Segmentation With Knowledge Distillation Focusing on Old Class (CVPR 2023 · similarity 0.88)
- ReBac: Current-to-Past Residual Background Correction for Class-Incremental Semantic Segmentation (IJCV 2026 · similarity 0.88)
- Revisiting Efficient Semantic Segmentation: Learning Offsets for Better Spatial and Class Feature Alignment (ICCV 2025 · similarity 0.88)
- Replay Without Saving: Prototype Derivation and Distribution Rebalance for Class-Incremental Semantic Segmentation (TPAMI 2025 · similarity 0.88)
- Cs2K: Class-specific and Class-shared Knowledge Guidance for Incremental Semantic Segmentation (ECCV 2024 · similarity 0.87)
- CoMBO: Conflict Mitigation via Branched Optimization for Class Incremental Segmentation (CVPR 2025 · similarity 0.85)
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.
