Research map / Continual learning / Class-incremental
Distillation-based continual segmentation
Preserves old classes in segmentation using similarity-weighted distillation, instance replay and memory selection.
17 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · an approach in Class-incremental, keyword cluster “segmentation · semantic · background”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
6 papers in 2023–24 and 6 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 6.5% to 4.4%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2021 · PLOP: Learning Without Forgetting for Continual Semantic Segmentation (CVPR 2021 · 265 citations)
- 2022 · Representation Compensation Networks for Continual Semantic Segmentation (CVPR 2022 · 116 citations)
- 2023 · Continual Semantic Segmentation With Automatic Memory Sample Selection (CVPR 2023 · 44 citations)
- 2024 · Continual Segmentation with Disentangled Objectness Learning and Class Recognition (CVPR 2024 · 15 citations)
- 2025 · Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation (CVPR 2025 · 4 citations)
- 2026 · Decoupling Continual Semantic Segmentation (AAAI 2026 · 1 citation)
Earliest papers here
- PLOP: Learning Without Forgetting for Continual Semantic Segmentation (CVPR 2021 · 265 citations)
- Continual Semantic Segmentation via Repulsion-Attraction of Sparse and Disentangled Latent Representations (CVPR 2021 · 152 citations)
- Representation Compensation Networks for Continual Semantic Segmentation (CVPR 2022 · 116 citations)
What has been tried
13 papers here report 157 measured changes (a component added, removed or swapped, compared with a baseline): helped 78%, no clear effect 4%, hurt 17%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add objectness-score distillation | PASCAL VOC 15-1 · all mIoU (%) | 18.44 -> 59.64 | Helped | Continual Segmentation with Disentangled Objectness Learning and Class Recognition (CVPR 2024 · 15 citations) |
| Replace Lkd with wrong-label knowledge distillation | Pascal VOC 2012, 15-1s · mIoU (all) | 54.68 -> 58.86 | Helped | TIKP: Text-to-Image Knowledge Preservation for Continual Semantic Segmentation (AAAI 2024 · 6 citations) |
| Add dynamic old-class combination | Pascal VOC 5-3 overlapped · mIoU | 58.52 -> 63.16 | Helped | Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation (CVPR 2025 · 4 citations) |
| Replace vanilla contrastive clustering loss with fairness contrastive clustering loss | ADE20K 100-50 · mIoU (all) | 34.8 -> 37.9 | Helped | FALCON: Fairness Learning via Contrastive Attention Approach to Continual Semantic Scene Understanding (CVPR 2025 · 4 citations) |
| Replace scratch-learned queries with lazy query pre-alignment | ADE20K 100-5 · mIoU, all classes | 13.2 -> 30.5 | Helped | Rethinking Query-based Transformer for Continual Image Segmentation (CVPR 2025 · 4 citations) |
| Add Ld to the attack objective without Lcs | Pascal VOC2012, 15-5 · all-class mIoU | 67.70 -> 63.08 | Hurt | Attacks on Continual Semantic Segmentation by Perturbing Incremental Samples (AAAI 2024 · 3 citations) |
| Add adversarial samples using AAKR attack to the PLOP baseline | 15-1s (6 tasks) · mIoU (all) | +8.29% | Helped | AAKR: Adversarial Attack-based Knowledge Retention for Continual Semantic Segmentation (AAAI 2025 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 5 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ADE20K · 4 of 5 papers
- Pascal VOC2012 · 2 of 5 papers
Baselines
- MiB · 4 of 5 papers
- PLOP · 4 of 5 papers
- ILT · 3 of 5 papers
- RCIL · 3 of 5 papers
- SSUL · 3 of 5 papers
- REMINDER · 2 of 5 papers
- EWF · 2 of 5 papers
- CoMFormer · 2 of 5 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 12 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 3 papers
- Hefei National Laboratory · 2 papers
- Google DeepMind · 2 papers
- University of Arkansas · 2 papers
Use it
Download BibTeX of 17 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
Elsewhere in the topic
- Exemplar-efficient class-incremental learning (111 papers)
- Class-balanced knowledge retention (73 papers)
- Few-shot class-incremental prototypes (47 papers)
- Incremental object detection with replay (32 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).