Research map / Label-efficient and robust learning / Semi-supervised learning
Consistency training for semi-supervised segmentation
Uses cross pseudo supervision and consistency with unreliable-label handling for segmentation.
49 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Semi-supervised learning, keyword cluster “pseudo labels · segmentation · pseudo labeling”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 3 tracked benchmarks.
14 papers in 2023–24 and 14 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 8.8% to 8.9%.
Open questions the data points to
- Cityscapes 1/16 (186) (mIoU (%)): best went from 76.81 (up to 2024) to 78.3 in 2025 — Beyond Confidence: Exploiting Homogeneous Pattern for Semi-Supervised Semantic Segmentation (ICML 2025)
No longer reported by papers here: Cityscapes val, 1/16 (186) (mIoU (%), last in 2024); Cityscapes val, 1/4 (744) (mIoU (%), last in 2024).
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 · Joint Word Segmentation, POS-Tagging and Syntactic Chunking (AAAI 2016 · 25 citations)
- 2019 · Unsupervised Learning Helps Supervised Neural Word Segmentation (AAAI 2019)
- 2020 · Semi-Supervised Semantic Segmentation With Cross-Consistency Training (CVPR 2020 · 981 citations)
- 2021 · Semi-Supervised Semantic Segmentation With Cross Pseudo Supervision (CVPR 2021 · 1,204 citations)
- 2022 · Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels (CVPR 2022 · 500 citations)
- 2023 · Augmentation Matters: A Simple-Yet-Effective Approach to Semi-Supervised Semantic Segmentation (CVPR 2023 · 147 citations)
- 2024 · Towards the Uncharted: Density-Descending Feature Perturbation for Semi-supervised Semantic Segmentation (CVPR 2024 · 41 citations)
- 2025 · Relaxed Class-consensus Consistency for Semi-supervised Semantic Segmentation (AAAI 2025 · 7 citations)
- 2026 · S5: Scalable Semi-Supervised Semantic Segmentation in Remote Sensing (AAAI 2026 · 2 citations)
Earliest papers here
- Joint Word Segmentation, POS-Tagging and Syntactic Chunking (AAAI 2016 · 25 citations)
- Unsupervised Learning Helps Supervised Neural Word Segmentation (AAAI 2019)
- Semi-Supervised Semantic Segmentation With Cross-Consistency Training (CVPR 2020 · 981 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
Cityscapes 1/16 (186) · mIoU (%) (higher is better)
6 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 76.81 | Switching Temporary Teachers for Semi-Supervised Semantic Segmentation (NeurIPS 2023 · 16 citations) |
| 2025 | 78.3 | Beyond Confidence: Exploiting Homogeneous Pattern for Semi-Supervised Semantic Segmentation (ICML 2025) |
Cityscapes val, 1/16 (186) · mIoU (%) (higher is better)
4 papers, 2022–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 74.9 | Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels (CVPR 2022 · 500 citations) |
| 2023 | 75.2 | Instance-Specific and Model-Adaptive Supervision for Semi-Supervised Semantic Segmentation (CVPR 2023 · 56 citations) |
| 2024 | 77.1 | Towards the Uncharted: Density-Descending Feature Perturbation for Semi-supervised Semantic Segmentation (CVPR 2024 · 41 citations) |
Cityscapes val, 1/4 (744) · mIoU (%) (higher is better)
4 papers, 2022–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 78.51 | Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels (CVPR 2022 · 500 citations) |
| 2024 | 79.88 | Towards the Uncharted: Density-Descending Feature Perturbation for Semi-supervised Semantic Segmentation (CVPR 2024 · 41 citations) |
What has been tried
40 papers here report 448 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 6%, hurt 19%, 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 density-descending feature perturbations instead of using image-level consistency alone | Pascal VOC 2012 classic, 1/4 (366) and 1/2 (732) · mIoU (%) | 75.44 -> 79.51; 77.56 -> 81.21 | Helped | Towards the Uncharted: Density-Descending Feature Perturbation for Semi-supervised Semantic Segmentation (CVPR 2024 · 41 citations) |
| add class-consensus consistency to UniMatch | PASCAL classic 1/16 (92) · mIoU | 67.4 -> 70.2 | Helped | Relaxed Class-consensus Consistency for Semi-supervised Semantic Segmentation (AAAI 2025 · 7 citations) |
| add feature alignment using MSE rather than the SWSEG loss | PASCAL VOC 2012 · mIoU | 70.2 vs 79.0 | Hurt | Improving Semi-Supervised Semantic Segmentation with Sliced-Wasserstein Feature Alignment and Uniformity (CVPR 2025 · 5 citations) |
| Add point erasure | SemanticKITTI, 1%, 10%, 20%, and 50% labeled data · mIoU | 45.4 -> 51.7; 56.1 -> 59.8; 57.8 -> 61.8; 58.7 -> 60.4 | Helped | Exploring Scene Affinity for Semi-Supervised LiDAR Semantic Segmentation (CVPR 2025 · 3 citations) |
| Add SAM top-1 IoU matching (IOM) | RefCOCO val, 1% labeled · oIoU | 42.62 -> 49.66 | Helped | SAM as the Guide: Mastering Pseudo-Label Refinement in Semi-Supervised Referring Expression Segmentation (ICML 2024 · 3 citations) |
| Add selective reliability-based multi-phase training instead of one-phase retraining | Pascal VOC 2012, 1/8 · MIoU | 72.1 -> 74.9 | Helped | RWMS: Reliable Weighted Multi-Phase for Semi-supervised Segmentation (AAAI 2024 · 2 citations) |
| Use collision cross-entropy instead of standard or reverse cross-entropy for soft pseudo-labels | PASCAL VOC 2012 val · mIoU | Helped | Soft Self-labeling and Potts Relaxations for Weakly-supervised Segmentation (CVPR 2025 · 1 citation) | |
| Use BCE alone instead of the full customized optimization criterion | Cityscapes fully supervised · accuracy | Hurt | Towards Robust Pseudo-Label Learning in Semantic Segmentation: An Encoding Perspective (NeurIPS 2025 · 0 citations) | |
| Increase the number of negative agents within the tested range | PASCAL VOC classic 1/16 (92), ResNet-50 · mIoU (%) | Helped | Beyond Confidence: Exploiting Homogeneous Pattern for Semi-Supervised Semantic Segmentation (ICML 2025) |
What reviewers will expect
Datasets and baselines named in the 7 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Cityscapes · 5 of 7 papers
- PASCAL VOC 2012 · 4 of 7 papers
- SBD · 2 of 7 papers
- ADE20K · 2 of 7 papers
- COCO · 2 of 7 papers
Baselines
- UniMatch · 4 of 7 papers
- FixMatch · 3 of 7 papers
- AugSeg · 3 of 7 papers
- DAW · 3 of 7 papers
- U2PL · 2 of 7 papers
- CPS · 2 of 7 papers
- GTA-Seg · 2 of 7 papers
- RankMatch · 2 of 7 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 28 papers; from OpenAlex and the papers' affiliations).
- The University of Sydney · 5 papers
- University of Science and Technology of China · 4 papers
- Zhejiang University · 2 papers
- Jilin University · 2 papers
- Baidu · 2 papers
- XJTLU · 2 papers
- Metavisioncn · 2 papers
- University of Liverpool · 2 papers
- Tsinghua University · 2 papers
Use it
Download BibTeX of 36 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
- Confidence-based pseudo-label selection (134 papers)
- Semi-supervised medical segmentation (50 papers)
- Pseudo-labeling for specialized tasks (48 papers)
Elsewhere in the topic
- Holistic semi-supervised pipelines (125 papers)
- Self-training theory (86 papers)
- Safe and realistic semi-supervised learning (70 papers)
- GAN-based semi-supervised learning (61 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).