Research map / Continual learning / Class-incremental
Continual Semantic Segmentation via Structure Preserving and Projected Feature Alignment
Zihan Lin, Zilei Wang, Yixin Zhang
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 |
|---|---|---|---|
| add pseudo labels for old classes to fine-tuning with cross-entropy (ablation) | Fine-tuning · VOC 2012 overlapped 15-1 | mIoU (all classes): 4.5 -> 19.5 | Improved |
| add projected feature alignment L_ali to the full framework (ablation) | SPPA · VOC 2012 overlapped 15-1 | mIoU (all classes): 53.8 -> 55.9 | Improved |
| add structure preserving loss L_str to the full framework (ablation) | SPPA · VOC 2012 overlapped 15-1 | mIoU (all classes): 51.2 -> 55.9 | Improved |
| use projected feature alignment instead of pooled distillation (ablation) | Lce + pseudo label baseline · VOC 2012 overlapped 15-1 | mIoU (all classes): 48.2 -> 50.5 | Improved |
| use projected feature alignment instead of pixel-wise feature alignment (ablation) | Lce + pseudo label baseline · VOC 2012 overlapped 15-1 | mIoU (all classes): 33.7 -> 50.5 | Improved |
| add feature space constraints (pixel-wise alignment) to the pseudo-label baseline (ablation) | Lce + pseudo label baseline · VOC 2012 overlapped 15-1 | mIoU (all classes): 18.9 -> 33.7 | Improved |
| use structure preserving loss instead of prototype alignment (ablation) | SPPA · VOC 2012 overlapped 15-1 | old-class discriminability (t-SNE) | Improved |
| SPPA versus PLOP with 10 classes per step | SPPA · ADE20K overlapped 100-10 | mIoU (all classes): 31.5 -> 31.5 | No clear effect |
| SPPA versus PLOP when adding five classes in one step | SPPA · VOC 2012 disjoint 15-5 | mIoU (all classes): 64.2 -> 69.0 | Improved |
Closest papers by meaning
- Continual Semantic Segmentation via Repulsion-Attraction of Sparse and Disentangled Latent Representations (CVPR 2021 · similarity 0.92)
- Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation (CVPR 2025 · similarity 0.91)
- TIKP: Text-to-Image Knowledge Preservation for Continual Semantic Segmentation (AAAI 2024 · similarity 0.90)
- Continual Segmentation with Disentangled Objectness Learning and Class Recognition (CVPR 2024 · similarity 0.88)
- Towards Continual Universal Segmentation (CVPR 2025 · similarity 0.88)
- Rethinking Query-based Transformer for Continual Image Segmentation (CVPR 2025 · 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.
