Research map / Detection and segmentation / CNN features
Boundary-aware instance segmentation
Refine instance masks with boundary and polygon modeling and top-down bottom-up blending.
83 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in CNN features, keyword cluster “instance · panoptic · mask”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
14 papers in 2023–24 and 13 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 4.3% to 3.7%.
No longer reported by papers here: Cityscapes val (AP, last in 2023); COCO test-dev (AP, last in 2023); Cityscapes test (AP, last in 2021); COCO val2017 (mask AP, last in 2021).
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 · Instance-Aware Semantic Segmentation via Multi-Task Network Cascades (CVPR 2016 · 1,294 citations)
- 2017 · Fully Convolutional Instance-Aware Semantic Segmentation (CVPR 2017 · 1,134 citations)
- 2018 · MaskLab: Instance Segmentation by Refining Object Detection With Semantic and Direction Features (CVPR 2018 · 405 citations)
- 2019 · Hybrid Task Cascade for Instance Segmentation (CVPR 2019 · 1,531 citations)
- 2020 · PointRend: Image Segmentation As Rendering (CVPR 2020 · 1,175 citations)
- 2021 · Boundary IoU: Improving Object-Centric Image Segmentation Evaluation (CVPR 2021 · 468 citations)
- 2022 · Masked-Attention Mask Transformer for Universal Image Segmentation (CVPR 2022 · 3,259 citations)
- 2023 · Mask DINO: Towards a Unified Transformer-Based Framework for Object Detection and Segmentation (CVPR 2023 · 543 citations)
- 2024 · Teeth-SEG: An Efficient Instance Segmentation Framework for Orthodontic Treatment based on Multi-Scale Aggregation and Anthropic Prior Knowledge (CVPR 2024 · 15 citations)
- 2025 · Efficient Connectivity-Preserving Instance Segmentation with Supervoxel-Based Loss Function (AAAI 2025 · 4 citations)
- 2026 · Disco: Densely-overlapping Cell Instance Segmentation via Adjacency-aware Collaborative Coloring (ICLR 2026 · 0 citations)
Earliest papers here
- Instance-Aware Semantic Segmentation via Multi-Task Network Cascades (CVPR 2016 · 1,294 citations)
- Instance-Level Segmentation for Autonomous Driving With Deep Densely Connected MRFs (CVPR 2016 · 215 citations)
- Iterative Instance Segmentation (CVPR 2016 · 148 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 test-dev · mask AP (higher is better)
13 papers, 2019–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 41.2 | Hybrid Task Cascade for Instance Segmentation (CVPR 2019 · 1,531 citations) |
| 2020 | 41.7 | SOLOv2: Dynamic and Fast Instance Segmentation (NeurIPS 2020 · 475 citations) |
| 2021 | 46.7 | SOLQ: Segmenting Objects by Learning Queries (NeurIPS 2021 · 57 citations) |
| 2022 | 49.2 | SOIT: Segmenting Objects with Instance-Aware Transformers (AAAI 2022 · 23 citations) |
Cityscapes val · AP (higher is better)
8 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 44.6 | PolyTransform: Deep Polygon Transformer for Instance Segmentation (CVPR 2020 · 188 citations) |
| 2021 | 46.9 | Look Closer To Segment Better: Boundary Patch Refinement for Instance Segmentation (CVPR 2021 · 99 citations) |
COCO test-dev · AP (higher is better)
6 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 41.3 | BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · 626 citations) |
| 2023 | 49.1 | CLUSTSEG: Clustering for Universal Segmentation (ICML 2023 · 26 citations) |
Cityscapes test · AP (higher is better)
6 papers, 2017–2021 · not reported since 2021
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 17.4 | Boundary-Aware Instance Segmentation (CVPR 2017 · 175 citations) |
| 2019 | 27.6 | Instance Segmentation by Jointly Optimizing Spatial Embeddings and Clustering Bandwidth (CVPR 2019 · 276 citations) |
| 2020 | 40.1 | PolyTransform: Deep Polygon Transformer for Instance Segmentation (CVPR 2020 · 188 citations) |
| 2021 | 42.7 | Look Closer To Segment Better: Boundary Patch Refinement for Instance Segmentation (CVPR 2021 · 99 citations) |
COCO test-dev · FPS (higher is better)
5 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 27.2 | Deep Snake for Real-Time Instance Segmentation (CVPR 2020 · 386 citations) |
| 2022 | 41.6 | Sparse Instance Activation for Real-Time Instance Segmentation (CVPR 2022 · 203 citations) |
| 2023 | 53.8 | FastInst: A Simple Query-Based Model for Real-Time Instance Segmentation (CVPR 2023 · 115 citations) |
COCO val2017 · mask AP (higher is better)
5 papers, 2020–2021 · not reported since 2021
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 40.9 | PointRend: Image Segmentation As Rendering (CVPR 2020 · 1,175 citations) |
| 2021 | 41.2 | DCT-Mask: Discrete Cosine Transform Mask Representation for Instance Segmentation (CVPR 2021 · 85 citations) |
What has been tried
78 papers here report 903 measured changes (a component added, removed or swapped, compared with a baseline): helped 78%, no clear effect 7%, hurt 14%, 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 |
|---|---|---|---|---|
| use permutation-based upscalers instead of bilinear interpolation | IO150K i.i.d. test · mIoU | 0.72 -> 0.91 | Helped | Teeth-SEG: An Efficient Instance Segmentation Framework for Orthodontic Treatment based on Multi-Scale Aggregation and Anthropic Prior Knowledge (CVPR 2024 · 15 citations) |
| Add the local instance decoder to a global-only decoder | COCO val2017 · AP | 28.7 -> 29.8 | Helped | MobileInst: Video Instance Segmentation on the Mobile (AAAI 2024 · 6 citations) |
| Use a positive α in the topological loss | Segmentation datasets · Topological accuracy | Helped | Efficient Connectivity-Preserving Instance Segmentation with Supervoxel-Based Loss Function (AAAI 2025 · 4 citations) | |
| Reduce input resolution to 32 × 64 | Cityscapes · IoU | 0.52 -> 0.36 | Hurt | Foveated Instance Segmentation (CVPR 2025 · 4 citations) |
| replace Centerness sample weights with RMask IoU | COCO val · AP | 29.1 -> 30.3 (+1.2) | Helped | PolarNeXt: Rethink Instance Segmentation with Polar Representation (CVPR 2025 · 3 citations) |
| replace the object decoder with direct query decoding | HQ-YTVIS val · AP | 51.4 -> 53.5 | Helped | Robust and Consistent Online Video Instance Segmentation via Instance Mask Propagation (AAAI 2025 · 1 citation) |
| add IoU-head adaptation | DAVIS-585 · Top-1 Acc | 51.5 / 62.1 | Helped | SAMRefiner: Taming Segment Anything Model for Universal Mask Refinement (ICLR 2025 · 1 citation) |
| add asymptotic training and encoding transformation | DSB2018 · DICE | 0.862 -> 0.883 | Helped | The Four Color Theorem for Cell Instance Segmentation (ICML 2025 · 0 citations) |
| Add SI-SCS to SCS | VOC 2012, 50% pruning · mask AP | 28.8 -> 32.4 | Helped | Training-Free Dataset Pruning for Instance Segmentation (ICLR 2025 · 0 citations) |
| Use 6 EFD decoder layers and 3 polygon decoder layers rather than fewer decoder layers | COCO val2017 · AP | 39.4 -> 40.6 | Helped | EFDTR: Learnable Elliptical Fourier Descriptor Transformer for Instance Segmentation (ICML 2025) |
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 |
|---|---|---|---|---|
| Mask R-CNN | COCO test-dev | mask AP | 35.6 – 38.8 | 6 papers, e.g. CVPR 2019 and NeurIPS 2021 |
| Mask R-CNN | COCO val2017 | AP | 34.7 – 36.5 | 3 papers, e.g. CVPR 2021 and NeurIPS 2020 |
| Mask2Former | COCO test-dev | AP | 38 – 47.9 | 2 papers, e.g. CVPR 2023 and ICML 2023 |
| PANet | Cityscapes test | AP | 31.8 – 36.4 | 2 papers, e.g. CVPR 2020 and CVPR 2020 |
| PolyTransform | Cityscapes val | AP | 40.2 – 44.6 | 2 papers, e.g. CVPR 2022 and CVPR 2021 |
What reviewers will expect
Datasets and baselines named in the 9 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 · 3 of 9 papers
- LVIS · 2 of 9 papers
- MS COCO · 2 of 9 papers
- PASCAL VOC 2012 · 2 of 9 papers
Baselines
- PolarMask · 2 of 9 papers
- Mask R-CNN · 2 of 9 papers
- Deep Snake · 2 of 9 papers
- DANCE · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 27 papers; from OpenAlex and the papers' affiliations).
- Hong Kong University of Science and Technology · 3 papers
- Tsinghua University · 3 papers
- The Hong Kong Polytechnic University · 2 papers
- Alibaba Group · 2 papers
- International Digital Economy Academy · 2 papers
- Shanghai Jiao Tong University · 2 papers
Use it
Download BibTeX of 35 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
- Joint detection and instance segmentation (39 papers)
- Unified panoptic segmentation (33 papers)
Elsewhere in the topic
- Context modeling for semantic segmentation (241 papers)
- Region-based and query-based detectors (183 papers)
- Local features and context in CNNs (170 papers)
- Context for hard detection conditions (133 papers)
- Domain-specific segmentation and detection (118 papers)
- Convolution-transformer backbones (101 papers)
- Box regression and loss design (94 papers)
- Efficient DETR query design (85 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).