Research map / Detection and segmentation / Weakly supervised detection and segmentation
CAM refinement and prototypes
Refine class activation maps with prototypes, SAM, and co-training for segmentation.
52 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in Weakly supervised detection and segmentation, keyword cluster “weakly supervised · supervised semantic · image level”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Leveling off Several papers from 2025–26 still report on 3 tracked benchmarks, but the best result has barely moved since 2024.
16 papers in 2023–24 and 13 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 13.9% to 18.1%.
Open questions the data points to
- Papers report Baseline on PASCAL VOC 2012 val (mIoU) anywhere from 47.2 to 68.7: comparisons against it depend on whose number you take.
What looks settled
- PASCAL VOC 2012 test (mIoU): best up to 2024 77.5, best in 2025–26 75.9 from 3 papers, at most 0.2% better, if at all.
- PASCAL VOC 2012 val (mIoU): best up to 2024 78.2, best in 2025–26 76 from 3 papers, at most 0.2% better, if at all.
- MS COCO 2014 val (mIoU): best up to 2024 49.8, best in 2025–26 47.2 from 2 papers, at most 0.2% better, if at all.
No longer reported by papers here: PASCAL VOC 2012 train (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
- 2020 · Self-Supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation (CVPR 2020 · 768 citations)
- 2021 · Embedded Discriminative Attention Mechanism for Weakly Supervised Semantic Segmentation (CVPR 2021 · 163 citations)
- 2022 · Multi-Class Token Transformer for Weakly Supervised Semantic Segmentation (CVPR 2022 · 297 citations)
- 2023 · Token Contrast for Weakly-Supervised Semantic Segmentation (CVPR 2023 · 207 citations)
- 2024 · From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · 80 citations)
- 2025 · FFR: Frequency Feature Rectification for Weakly Supervised Semantic Segmentation (CVPR 2025 · 15 citations)
- 2026 · CaPro: Curvilinear-aware Prompt Learning with Single Unlabeled Image for Cost-effective Curvilinear Structure Segmentation (AAAI 2026 · 2 citations)
Earliest papers here
- Self-Supervised Equivariant Attention Mechanism for Weakly Supervised Semantic Segmentation (CVPR 2020 · 768 citations)
- Weakly-Supervised Semantic Segmentation via Sub-Category Exploration (CVPR 2020 · 319 citations)
- Learning Integral Objects With Intra-Class Discriminator for Weakly-Supervised Semantic Segmentation (CVPR 2020 · 219 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.
PASCAL VOC 2012 test · mIoU (higher is better)
27 papers, 2020–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 65.9 | Weakly-Supervised Semantic Segmentation via Sub-Category Exploration (CVPR 2020 · 319 citations) |
| 2021 | 70.6 | Embedded Discriminative Attention Mechanism for Weakly Supervised Semantic Segmentation (CVPR 2021 · 163 citations) |
| 2022 | 72.9 | Towards Noiseless Object Contours for Weakly Supervised Semantic Segmentation (CVPR 2022 · 41 citations) |
| 2024 | 77.5 | From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · 80 citations) |
PASCAL VOC 2012 val · mIoU (higher is better)
27 papers, 2020–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 66.1 | Weakly-Supervised Semantic Segmentation via Sub-Category Exploration (CVPR 2020 · 319 citations) |
| 2021 | 70.9 | Embedded Discriminative Attention Mechanism for Weakly Supervised Semantic Segmentation (CVPR 2021 · 163 citations) |
| 2022 | 72 | Towards Noiseless Object Contours for Weakly Supervised Semantic Segmentation (CVPR 2022 · 41 citations) |
| 2023 | 73.7 | Boundary-Enhanced Co-Training for Weakly Supervised Semantic Segmentation (CVPR 2023 · 107 citations) |
| 2024 | 78.2 | From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · 80 citations) |
MS COCO 2014 val · mIoU (higher is better)
19 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 44.7 | Threshold Matters in WSSS: Manipulating the Activation for the Robust and Accurate Segmentation Model Against Thresholds (CVPR 2022 · 96 citations) |
| 2023 | 45.3 | Weakly Supervised Semantic Segmentation via Adversarial Learning of Classifier and Reconstructor (CVPR 2023 · 88 citations) |
| 2024 | 49.8 | From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · 80 citations) |
PASCAL VOC 2012 train · mIoU (higher is better)
8 papers, 2022–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 66 | C2AM: Contrastive Learning of Class-Agnostic Activation Map for Weakly Supervised Object Localization and Semantic Segmentation (CVPR 2022) |
| 2023 | 72.3 | Weakly Supervised Semantic Segmentation via Adversarial Learning of Classifier and Reconstructor (CVPR 2023 · 88 citations) |
| 2024 | 81.7 | From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · 80 citations) |
What has been tried
44 papers here report 512 measured changes (a component added, removed or swapped, compared with a baseline): helped 75%, no clear effect 5%, hurt 16%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add logit-level SSC | PASCAL VOC 2012 train · mIoU (%) | 50.1 -> 57.3 | Helped | From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · 80 citations) |
| Add plain context prototype learning | PASCAL VOC 2012 train · mIoU | 50.1 -> 51.2 | Helped | Hunting Attributes: Context Prototype-Aware Learning for Weakly Supervised Semantic Segmentation (CVPR 2024 · 54 citations) |
| Add the background token and background CAM loss | PASCAL VOC 2012 train · mIoU | 64.7 -> 67.6 | Helped | Class Tokens Infusion for Weakly Supervised Semantic Segmentation (CVPR 2024 · 51 citations) |
| add dual student architecture with discrepancy loss | PASCAL VOC 2012 val · CAM mIoU | 63.2 -> 67.3 with DS + Ldis (approx. +4.1 cumulative with discrepancy) | Helped | DuPL: Dual Student with Trustworthy Progressive Learning for Robust Weakly Supervised Semantic Segmentation (CVPR 2024 · 46 citations) |
| Add prototype CAM to DeepLab V2 output | PASCAL VOC 2012 train · mIoU | 59.7 -> 61.3 | Helped | PSDPM: Prototype-based Secondary Discriminative Pixels Mining for Weakly Supervised Semantic Segmentation (CVPR 2024 · 24 citations) |
| Add the feature rectification loss without mask Bl | PASCAL VOC 2012 val · mIoU (segmentation) | 72.1 -> 70.4 | Hurt | FFR: Frequency Feature Rectification for Weakly Supervised Semantic Segmentation (CVPR 2025 · 15 citations) |
| add foreground and background matching losses rather than using classification loss alone | BCSS-WSSS initial pseudo masks · mIoU (%) | 62.08±0.3 -> 67.54±0.6 | Helped | Prototype-Based Image Prompting for Weakly Supervised Histopathological Image Segmentation (CVPR 2025 · 14 citations) |
| Add Patch-level Confidence Allocation (PCA) | PASCAL VOC 2012 val · mIoU | 70.7 -> 72.1 | Helped | Weakly Supervised Semantic Segmentation via Progressive Confidence Region Expansion (CVPR 2025 · 14 citations) |
| Add CARB to the base method | Cityscapes val · mIoU | 40.1 -> 45.7 | Helped | Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 2024 · 14 citations) |
| add L_cre supervision | PASCAL VOC val · mIoU | 65.0 -> 76.4 | Helped | MoRe: Class Patch Attention Needs Regularization for Weakly Supervised Semantic Segmentation (AAAI 2025 · 12 citations) |
| Prioritize uncertain patches for masking instead of random patches | PASCAL VOC 2012 val · mIoU | random: 73.1, 73.6, 74.1, 74.2, 73.2; uncertain: 74.1, 74.2, 73.9, 74.4, 73.7 | Mixed | Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · 11 citations) |
| Add CCA | ProG · mIoU | 80.01±0.46 -> 82.90±0.10 | Helped | Weakly Supervised Gland Segmentation with Class Semantic Consistency and Purified Labels Filtration (AAAI 2025 · 7 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| Baseline | PASCAL VOC 2012 val | mIoU | 47.2 – 68.7 | 3 papers, e.g. AAAI 2023 and CVPR 2023 |
| FickleNet | PASCAL VOC 2012 val | mIoU | 61.2 – 64.9 | 3 papers, e.g. AAAI 2020 and CVPR 2020 |
| AffinityNet | PASCAL VOC 2012 val | mIoU | 58.4 – 61.7 | 3 papers, e.g. AAAI 2020 and CVPR 2020 |
| FickleNet | PASCAL VOC 2012 test | mIoU | 61.9 – 65.3 | 3 papers, e.g. AAAI 2020 and CVPR 2020 |
| MCTformer+ | MS COCO 2014 val | mIoU | 42 – 45.2 | 2 papers, e.g. CVPR 2023 and AAAI 2024 |
What reviewers will expect
Datasets and baselines named in the 8 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- PASCAL VOC 2012 · 4 of 8 papers
- MS COCO 2014 · 4 of 8 papers
- SBD · 2 of 8 papers
Baselines
- CLIP-ES · 4 of 8 papers
- ToCo · 4 of 8 papers
- DuPL · 4 of 8 papers
- CLIMS · 4 of 8 papers
- MCTformer · 3 of 8 papers
- AFA · 3 of 8 papers
- ReCAM · 2 of 8 papers
- MLPS · 2 of 8 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 29 papers; from OpenAlex and the papers' affiliations).
- Korea Advanced Institute of Science and Technology · 4 papers
- Tsinghua University · 2 papers
- Xidian University · 2 papers
- Fudan University · 2 papers
- XJTLU · 2 papers
- University of Liverpool · 2 papers
- Shenzhen University · 2 papers
Use it
Download BibTeX of 40 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
- Weakly supervised semantic segmentation (116 papers)
- Weakly supervised detection (97 papers)
- Weakly supervised object localization (54 papers)
Elsewhere in the topic
- Pseudo-labeling for semi-supervised detection (76 papers)
- Consistency regularization for segmentation (25 papers)
- Semi-supervised instance segmentation (16 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).