Research map / Detection and segmentation / Open-set and few-shot detection
Few-shot segmentation
Segment novel classes from few support images using pretrained features and aggregation.
64 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Open-set and few-shot detection, keyword cluster “zero · shot object · shot segmentation”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 5 of 6 tracked benchmarks.
23 papers in 2023–24 and 11 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 3.9% to 2.0%.
Open questions the data points to
- COCO-20i 1-shot (mIoU): best went from 50 (up to 2024) to 71.33 in 2025 — DSV-LFS: Unifying LLM-Driven Semantic Cues with Visual Features for Robust Few-Shot Segmentation (CVPR 2025 · 6 citations)
- COCO-20i 5-shot (mIoU): best went from 56 (up to 2024) to 71.9 in 2025 — DSV-LFS: Unifying LLM-Driven Semantic Cues with Visual Features for Robust Few-Shot Segmentation (CVPR 2025 · 6 citations)
- PASCAL-5i 1-shot (mIoU): best went from 72.1 (up to 2024) to 74.96 in 2025 — DSV-LFS: Unifying LLM-Driven Semantic Cues with Visual Features for Robust Few-Shot Segmentation (CVPR 2025 · 6 citations)
- PASCAL-5i 5-shot (mIoU): best went from 72.1 (up to 2024) to 75.21 in 2025 — DSV-LFS: Unifying LLM-Driven Semantic Cues with Visual Features for Robust Few-Shot Segmentation (CVPR 2025 · 6 citations)
- COCO-20ᵢ 1-shot (mIoU): best went from 52.7 (up to 2024) to 60.2 in 2025 — SANSA: Unleashing the Hidden Semantics in SAM2 for Few-Shot Segmentation (NeurIPS 2025 · 4 citations)
- Papers report Baseline on PASCAL-5i 1-shot (mIoU) anywhere from 53 to 59.3: comparisons against it depend on whose number you take.
No longer reported by papers here: PASCAL-5i 1-shot (FB-IoU, 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
- 2018 · One-Shot Segmentation in Clutter (ICML 2018 · 2 citations)
- 2019 · CANet: Class-Agnostic Segmentation Networks With Iterative Refinement and Attentive Few-Shot Learning (CVPR 2019 · 688 citations)
- 2020 · FSS-1000: A 1000-Class Dataset for Few-Shot Segmentation (CVPR 2020 · 255 citations)
- 2021 · Adaptive Prototype Learning and Allocation for Few-Shot Segmentation (CVPR 2021 · 469 citations)
- 2022 · GroupViT: Semantic Segmentation Emerges From Text Supervision (CVPR 2022 · 443 citations)
- 2023 · Hierarchical Dense Correlation Distillation for Few-Shot Segmentation (CVPR 2023 · 167 citations)
- 2024 · pix2gestalt: Amodal Segmentation by Synthesizing Wholes (CVPR 2024 · 38 citations)
- 2025 · Foreground-Covering Prototype Generation and Matching for SAM-Aided Few-Shot Segmentation (AAAI 2025 · 11 citations)
- 2026 · gen2seg: Generative Models Enable Generalizable Instance Segmentation (ICLR 2026 · 0 citations)
Earliest papers here
- One-Shot Segmentation in Clutter (ICML 2018 · 2 citations)
- CANet: Class-Agnostic Segmentation Networks With Iterative Refinement and Attentive Few-Shot Learning (CVPR 2019 · 688 citations)
- Semantic Projection Network for Zero- and Few-Label Semantic Segmentation (CVPR 2019 · 245 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.
COCO-20i 1-shot · mIoU (higher is better)
12 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 40.3 | Few-Shot Segmentation via Cycle-Consistent Transformer (NeurIPS 2021 · 31 citations) |
| 2022 | 44.2 | Mask Matching Transformer for Few-Shot Segmentation (NeurIPS 2022 · 20 citations) |
| 2023 | 50 | Hierarchical Dense Correlation Distillation for Few-Shot Segmentation (CVPR 2023 · 167 citations) |
| 2025 | 71.33 | DSV-LFS: Unifying LLM-Driven Semantic Cues with Visual Features for Robust Few-Shot Segmentation (CVPR 2025 · 6 citations) |
COCO-20i 5-shot · mIoU (higher is better)
12 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 45.6 | Few-Shot Segmentation via Cycle-Consistent Transformer (NeurIPS 2021 · 31 citations) |
| 2022 | 49.8 | Dynamic Prototype Convolution Network for Few-Shot Semantic Segmentation (CVPR 2022 · 119 citations) |
| 2023 | 56 | Hierarchical Dense Correlation Distillation for Few-Shot Segmentation (CVPR 2023 · 167 citations) |
| 2025 | 71.9 | DSV-LFS: Unifying LLM-Driven Semantic Cues with Visual Features for Robust Few-Shot Segmentation (CVPR 2025 · 6 citations) |
PASCAL-5i 1-shot · mIoU (higher is better)
12 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 61.2 | Attention-Based Multi-Context Guiding for Few-Shot Semantic Segmentation (AAAI 2019 · 145 citations) |
| 2021 | 64 | Few-Shot Segmentation via Cycle-Consistent Transformer (NeurIPS 2021 · 31 citations) |
| 2022 | 66.8 | Intermediate Prototype Mining Transformer for Few-Shot Semantic Segmentation (NeurIPS 2022 · 37 citations) |
| 2023 | 69.4 | Hierarchical Dense Correlation Distillation for Few-Shot Segmentation (CVPR 2023 · 167 citations) |
| 2024 | 72.1 | Bidirectional Reciprocative Information Communication for Few-Shot Semantic Segmentation (ICML 2024) |
| 2025 | 74.96 | DSV-LFS: Unifying LLM-Driven Semantic Cues with Visual Features for Robust Few-Shot Segmentation (CVPR 2025 · 6 citations) |
PASCAL-5i 5-shot · mIoU (higher is better)
12 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 62.2 | Attention-Based Multi-Context Guiding for Few-Shot Semantic Segmentation (AAAI 2019 · 145 citations) |
| 2021 | 67.5 | Few-Shot Segmentation via Cycle-Consistent Transformer (NeurIPS 2021 · 31 citations) |
| 2022 | 69.9 | Dynamic Prototype Convolution Network for Few-Shot Semantic Segmentation (CVPR 2022 · 119 citations) |
| 2023 | 71.8 | Hierarchical Dense Correlation Distillation for Few-Shot Segmentation (CVPR 2023 · 167 citations) |
| 2024 | 72.1 | Bidirectional Reciprocative Information Communication for Few-Shot Semantic Segmentation (ICML 2024) |
| 2025 | 75.21 | DSV-LFS: Unifying LLM-Driven Semantic Cues with Visual Features for Robust Few-Shot Segmentation (CVPR 2025 · 6 citations) |
PASCAL-5i 1-shot · FB-IoU (higher is better)
6 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 66.2 | CANet: Class-Agnostic Segmentation Networks With Iterative Refinement and Attentive Few-Shot Learning (CVPR 2019 · 688 citations) |
| 2021 | 73 | Few-Shot Segmentation via Cycle-Consistent Transformer (NeurIPS 2021 · 31 citations) |
| 2022 | 77.1 | Intermediate Prototype Mining Transformer for Few-Shot Semantic Segmentation (NeurIPS 2022 · 37 citations) |
| 2024 | 80.51 | CGMGM: A Cross-Gaussian Mixture Generative Model for Few-Shot Semantic Segmentation (AAAI 2024 · 11 citations) |
COCO-20ᵢ 1-shot · mIoU (higher is better)
4 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 32.56 | Anti-Aliasing Semantic Reconstruction for Few-Shot Semantic Segmentation (CVPR 2021 · 56 citations) |
| 2024 | 52.7 | Adaptive FSS: A Novel Few-Shot Segmentation Framework via Prototype Enhancement (AAAI 2024 · 16 citations) |
| 2025 | 60.2 | SANSA: Unleashing the Hidden Semantics in SAM2 for Few-Shot Segmentation (NeurIPS 2025 · 4 citations) |
What has been tried
56 papers here report 641 measured changes (a component added, removed or swapped, compared with a baseline): helped 77%, no clear effect 4%, hurt 17%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add the unlabeled branch | 1-shot, Res-50 · mIoU | 64.0 -> 66.1 | Helped | Relevant Intrinsic Feature Enhancement Network for Few-Shot Semantic Segmentation (AAAI 2024 · 29 citations) |
| add Support Modulation (SM) | PASCAL-5i fold-0 · mIoU | 61.48 -> 68.20 | Helped | Addressing Background Context Bias in Few-Shot Segmentation through Iterative Modulation (CVPR 2024 · 26 citations) |
| freeze the encoder and alternately train the two decoders instead of joint training | Mixed dataset · Registration Dice (%) | 79.1 ± 2.0 -> 79.8 ± 2.0; 0.7% | Helped | Bi-level Learning of Task-Specific Decoders for Joint Registration and One-Shot Medical Image Segmentation (CVPR 2024 · 23 citations) |
| add classifier-free guided mask sampling | COCO-Novel, 1-shot · segmentation AP | 6.07 -> 6.26 | Helped | MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · 17 citations) |
| add QUEST | S3DIS, 2-way 1-shot, S0 · mIoU | 47.32 -> 64.84 | Helped | No Time to Train: Empowering Non-Parametric Networks for Few-shot 3D Scene Segmentation (CVPR 2024 · 17 citations) |
| Add masked-pooling prompt initialization instead of random initialization, without causal attention | COCO-20i 1-shot · mean mIoU | 30.70 -> 30.87 | Helped | Visual Prompting for Generalized Few-shot Segmentation: A Multi-scale Approach (CVPR 2024 · 17 citations) |
| Add LAM to the unadapted FPTrans baseline | PASCAL-5ᵢ · mIoU | 68.8 -> 70.1 | Helped | Adaptive FSS: A Novel Few-Shot Segmentation Framework via Prototype Enhancement (AAAI 2024 · 16 citations) |
| add self Mamba blocks to the basic model | PASCAL-5ᵢ 1-shot · mIoU | 65.2 -> 67.4 | Helped | Hybrid Mamba for Few-Shot Segmentation (NeurIPS 2024 · 16 citations) |
| Add query-relevant scoring without target-relevant scoring | PASCAL-5ᵢ fold-1, 1-shot · mIoU | 73.8 -> 73.4 | Hurt | Task-Disruptive Background Suppression for Few-Shot Segmentation (AAAI 2024 · 12 citations) |
| Use SAM features instead of ResNet features in the prototype-pixel baseline | PASCAL-5^i, ResNet-50 · mIoU | 71.8 -> 66.2 | Hurt | Foreground-Covering Prototype Generation and Matching for SAM-Aided Few-Shot Segmentation (AAAI 2025 · 11 citations) |
| Add CDGMG&QCPE alone | PASCAL-5i 1-shot · mIoU (%) | 67.81 -> 66.95 | Hurt | CGMGM: A Cross-Gaussian Mixture Generative Model for Few-Shot Semantic Segmentation (AAAI 2024 · 11 citations) |
| Train with randomly selected 1 to N support samples | COCO-20^i strict 5-shot mean · mIoU | 56.0 -> 60.7 | Helped | Unleashing the Potential of the Diffusion Model in Few-shot Semantic Segmentation (NeurIPS 2024 · 11 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-5i 1-shot | mIoU | 53 – 59.3 | 2 papers, e.g. AAAI 2019 and NeurIPS 2021 |
What reviewers will expect
Datasets and baselines named in the 6 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 6 papers
- COCO-20i · 3 of 6 papers
- PASCAL-5i · 2 of 6 papers
Baselines
- HDMNet · 2 of 6 papers
- VRP-SAM · 2 of 6 papers
- SSP · 2 of 6 papers
- BAM · 2 of 6 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 34 papers; from OpenAlex and the papers' affiliations).
- Zhejiang University · 4 papers
- University of Science and Technology of China · 3 papers
- Sungkyunkwan University · 2 papers
- The Chinese University of Hong Kong · 2 papers
- Singapore University of Technology and Design · 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
- Few-shot object detection (67 papers)
- Zero-shot segmentation and detection (49 papers)
- Prompted SAM segmentation (39 papers)
Elsewhere in the topic
- Object-level representation learning (219 papers)
- Interactive and referring segmentation (161 papers)
- Open-set and unknown object detection (160 papers)
- Amodal and open-world instance segmentation (137 papers)
- Industrial anomaly detection (117 papers)
- Unsupervised object and part discovery (114 papers)
- Domain-adaptive object detection (103 papers)
- Open-vocabulary object detection (96 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).