Research map / Continual learning / Class-incremental
A Survey on Continual Semantic Segmentation: Theory, Challenge, Method and Application
Bo Yuan, Danpei Zhao
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 exemplar-memory replay to SSUL | SSUL · VOC 2012 10-1 overlapped (all classes) | mIoU (%): 58.23 -> 65.45 | Improved |
| add multi-level knowledge distillation (PLOP instead of MiB) | MiB · VOC 2012 10-1 overlapped (all classes) | mIoU (%): 12.65 -> 30.45 | Improved |
| add asymmetric region-wise contrastive learning to multi-level KD (IDEC) | multi-level KD · VOC 2012 10-1 overlapped (all classes) | mIoU (%): 30.45 -> 59.10 | Improved |
| use a stronger segmentation model (ViT-B/16 Incrementer instead of DeepLabv3 MiB) | MiB · VOC 2012 10-1 overlapped (all classes) | mIoU (%): 12.65 -> 70.16 | Improved |
| use SegFormer-B2 with replay (SATS-M instead of SATS) | SATS · VOC 2012 10-1 overlapped (all classes) | mIoU (%): 61.60 -> 69.27 | Improved |
| add AWT to MiB | MiB · VOC 2012 10-1 overlapped (all classes) | mIoU (%): 12.65 -> 26.00 | Improved |
| use a stronger segmentation model (Incrementer ViT-B/16 instead of MiB) | MiB · ADE20K 100-10 overlapped (all classes) | mIoU (%): 29.24 -> 43.85 | Improved |
| add replay (SSUL-M instead of data-free MiB) | MiB · VOC 2012 15-5 overlapped (all classes) | mIoU (%): 70.08 -> 73.19 | Improved |
| use continual learning methods instead of plain fine-tuning | fine-tuning · Cityscapes 11-5 / 1-1 | mIoU (%): 1-1: 42.9 vs MiB 42.2 | No clear effect |
| add distillation-based CSS (MiB) over fine-tuning on Cityscapes 1-1 | fine-tuning · Cityscapes 1-1 (21 steps) | mIoU (%): 42.9 -> 42.2 | Hurt |
| add RCIL over MiB on Cityscapes | MiB · Cityscapes 1-1 (21 steps) | mIoU (%): 42.2 -> 48.9 | Improved |
| add VOC 15-1 incremental KD (MiB) versus fine-tuning | fine tuning · VOC 2012 15-1 overlapped (all classes) | mIoU (%): 0.60 -> 29.29 | Improved |
Closest papers by meaning
- Preparing the Future for Continual Semantic Segmentation (ICCV 2023 · similarity 0.94)
- Learning at a Glance: Towards Interpretable Data-Limited Continual Semantic Segmentation via Semantic-Invariance Modelling (TPAMI 2024 · similarity 0.93)
- Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation (CVPR 2025 · similarity 0.92)
- Towards Continual Universal Segmentation (CVPR 2025 · similarity 0.91)
- SimCS: Simulation for Domain Incremental Online Continual Segmentation (AAAI 2024 · similarity 0.90)
- TIKP: Text-to-Image Knowledge Preservation for Continual Semantic Segmentation (AAAI 2024 · similarity 0.89)
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.
