Research map / Continual learning / Class-incremental
Unknown-class mining for incremental segmentation
Models background shift by mining unseen classes and using exemplar or prototype replay.
23 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · 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 10 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 6.5% to 7.4%.
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
- 2016 · Knowledge Transfer with Interactive Learning of Semantic Relationships (AAAI 2016 · 8 citations)
- 2020 · Modeling the Background for Incremental Learning in Semantic Segmentation (CVPR 2020 · 344 citations)
- 2021 · Class-Incremental Instance Segmentation via Multi-Teacher Networks (AAAI 2021 · 25 citations)
- 2022 · Mining Unseen Classes via Regional Objectness: A Simple Baseline for Incremental Segmentation (NeurIPS 2022 · 12 citations)
- 2023 · Federated Incremental Semantic Segmentation (CVPR 2023 · 58 citations)
- 2024 · Incremental Nuclei Segmentation from Histopathological Images via Future-class Awareness and Compatibility-inspired Distillation (CVPR 2024 · 11 citations)
- 2025 · Adaptive Prototype Replay for Class Incremental Semantic Segmentation (AAAI 2025 · 10 citations)
- 2026 · CiNuSeg: Class Incremental Nuclei Segmentation via Anchor-driven Consistency Learning with Dual Region Regularization (AAAI 2026 · 0 citations)
Earliest papers here
- Knowledge Transfer with Interactive Learning of Semantic Relationships (AAAI 2016 · 8 citations)
- Modeling the Background for Incremental Learning in Semantic Segmentation (CVPR 2020 · 344 citations)
- Class-Incremental Instance Segmentation via Multi-Teacher Networks (AAAI 2021 · 25 citations)
What has been tried
17 papers here report 178 measured changes (a component added, removed or swapped, compared with a baseline): helped 78%, no clear effect 5%, hurt 13%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add future-class awareness to PLOP + EWF | MoNuSAC 1-1 (4 tasks) · mDice (Mean) | 67.06 -> 68.60 | Helped | Incremental Nuclei Segmentation from Histopathological Images via Future-class Awareness and Compatibility-inspired Distillation (CVPR 2024 · 11 citations) |
| Add ADC to fixed prototype replay baseline | PASCAL VOC overlapped 15-1 (6 steps) · mIoU, all classes | 71.3 -> 71.9 | Helped | Adaptive Prototype Replay for Class Incremental Semantic Segmentation (AAAI 2025 · 10 citations) |
| Add HDHL to vanilla losses with pseudo-labeling | ADE20K CIPS 100-10 (6 steps) · PQ | 32.2 -> 34.3 | Helped | CoMBO: Conflict Mitigation via Branched Optimization for Class Incremental Segmentation (CVPR 2025 · 3 citations) |
| Add ITA to the KD-only model | S3DIS (Cnovel=5, M0) and ScanNet (Cnovel=5, M1) · mIoU (%) | 43.5 -> 41.4; 19.4 -> 32.6 | Mixed | LAGD: Local Topological-Alignment and Global Semantic-Deconstruction for Incremental 3D Semantic Segmentation (AAAI 2025 · 2 citations) |
| Learn before forgetting rather than forget before learning | ImageNet-100; four learning sets, 5- and 10-class settings · Acc_f | 80.8 -> 0; 70.8 -> 0; 77.2 -> 0; 74.8 -> 0; 82 -> 0; 82 -> 0.6; 83 -> 0.2; 82.2 -> 0 | Helped | Knowledge Swapping via Learning and Unlearning (ICML 2025 · 0 citations) |
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 · 3 of 5 papers
Baselines
- MiB · 2 of 5 papers
- PLOP · 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 16 papers; from OpenAlex and the papers' affiliations).
- Huazhong University of Science and Technology · 2 papers
- Beijing Jiaotong University · 2 papers
Use it
Download BibTeX of 23 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
- Distillation-based continual segmentation (17 papers)
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).