Research map / Video / Video segmentation and tracking
Matching-based video object segmentation
Propagates object masks via feature matching, instance queries and efficient linear attention.
86 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Video segmentation and tracking, keyword cluster “object segmentation · video object · vos”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 2 tracked benchmarks.
16 papers in 2023–24 and 41 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.6% to 7.6%.
Open questions the data points to
- DAVIS 2017 val (J&F): best went from 88.2 (up to 2024) to 90.7 in 2025 — SAM 2: Segment Anything in Images and Videos (ICLR 2025 · 264 citations)
What looks settled
- Ref-YouTube-VOS (J&F): best up to 2024 68.4, best in 2025–26 64.3 from 2 papers, at most 0.2% better, if at all.
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 · A Benchmark Dataset and Evaluation Methodology for Video Object Segmentation (CVPR 2016 · 2,095 citations)
- 2017 · One-Shot Video Object Segmentation (CVPR 2017 · 926 citations)
- 2018 · Automatic Segmentation of Data Sequences (AAAI 2018 · 10 citations)
- 2019 · FEELVOS: Fast End-To-End Embedding Learning for Video Object Segmentation (CVPR 2019 · 463 citations)
- 2020 · State-Aware Tracker for Real-Time Video Object Segmentation (CVPR 2020 · 124 citations)
- 2021 · End-to-End Video Instance Segmentation With Transformers (CVPR 2021 · 654 citations)
- 2022 · Implicit Motion Handling for Video Camouflaged Object Detection (CVPR 2022 · 98 citations)
- 2023 · Look Before You Match: Instance Understanding Matters in Video Object Segmentation (CVPR 2023 · 49 citations)
- 2024 · MemSAM: Taming Segment Anything Model for Echocardiography Video Segmentation (CVPR 2024 · 62 citations)
- 2025 · SAM 2: Segment Anything in Images and Videos (ICLR 2025 · 264 citations)
- 2026 · Open-World Object Counting in Videos (AAAI 2026 · 3 citations)
Earliest papers here
- A Benchmark Dataset and Evaluation Methodology for Video Object Segmentation (CVPR 2016 · 2,095 citations)
- Video Segmentation via Object Flow (CVPR 2016 · 344 citations)
- Automatic Fence Segmentation in Videos of Dynamic Scenes (CVPR 2016 · 23 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.
DAVIS 2017 val · J&F (higher is better)
5 papers, 2020–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 72.3 | State-Aware Tracker for Real-Time Video Object Segmentation (CVPR 2020 · 124 citations) |
| 2023 | 88.2 | Look Before You Match: Instance Understanding Matters in Video Object Segmentation (CVPR 2023 · 49 citations) |
| 2025 | 90.7 | SAM 2: Segment Anything in Images and Videos (ICLR 2025 · 264 citations) |
Ref-YouTube-VOS · J&F (higher is better)
4 papers, 2024–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 68.4 | Referred by Multi-Modality: A Unified Temporal Transformer for Video Object Segmentation (AAAI 2024 · 40 citations) |
What has been tried
58 papers here report 656 measured changes (a component added, removed or swapped, compared with a baseline): helped 75%, no clear effect 5%, hurt 20%, 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 Hiera-L instead of Hiera-B+ | MOSE val · J&F | 76.6 -> 77.9 | Helped | SAM 2: Segment Anything in Images and Videos (ICLR 2025 · 264 citations) |
| add the initial point prompt | CAMUS-Semi · mDice | 93.17 -> 93.31 | No clear effect | MemSAM: Taming Segment Anything Model for Echocardiography Video Segmentation (CVPR 2024 · 62 citations) |
| add sentence decoupling | MeViS · J &F | 39.7 -> 42.5 | Helped | Decoupling Static and Hierarchical Motion Perception for Referring Video Segmentation (CVPR 2024 · 41 citations) |
| add MTA and MTI | Ref-YouTube-VOS · J&F | 60.2% to 61.9% | Helped | Referred by Multi-Modality: A Unified Temporal Transformer for Video Object Segmentation (AAAI 2024 · 40 citations) |
| restrict memory banks | VOST · Jtr | 37.0 -> 38.6 | Helped | RMem: Restricted Memory Banks Improve Video Object Segmentation (CVPR 2024 · 31 citations) |
| Add IMA with prototype embedding to the baseline | DAVIS 2016 validation set · GM | 83.4 -> 85.9 | Helped | Dual Prototype Attention for Unsupervised Video Object Segmentation (CVPR 2024 · 25 citations) |
| add event input to image input | LLE-DAVIS · J &F | 0.559 -> 0.628 | Helped | Event-assisted Low-Light Video Object Segmentation (CVPR 2024 · 24 citations) |
| Add prompt features as decoder queries | YouTube-VIS 2021 development set · mAP | 45.9 -> 39.9 | Hurt | UniVS: Unified and Universal Video Segmentation with Prompts as Queries (CVPR 2024 · 22 citations) |
| Add short-text prediction branch | A2D-Sentences · mAP | 45.4 -> 47.8 | Helped | LoSh: Long-Short Text Joint Prediction Network for Referring Video Object Segmentation (CVPR 2024 · 18 citations) |
| add semantic feature alignment | MeViS Val · J&F | 42.4 -> 46.3 | Helped | Semantic and Sequential Alignment for Referring Video Object Segmentation (CVPR 2025 · 15 citations) |
| Add neighboring frames (motion guidance) to TPAVI | AVSBench-S4 · mIoU (MJ) | 72.80 -> 76.17 | Helped | Audio-Visual Segmentation via Unlabeled Frame Exploitation (CVPR 2024 · 13 citations) |
| Add the 2D Spatial Perceiver memory compressor | SA-V val · J&F | 63.5 -> 64.4 | Helped | EdgeTAM: On-Device Track Anything Model (CVPR 2025 · 11 citations) |
What reviewers will expect
Datasets and baselines named in the 17 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- SA-V · 4 of 17 papers
- SA-1B · 3 of 17 papers
- MOSE · 3 of 17 papers
- MeViS · 2 of 17 papers
- Ref-YouTube-VOS · 2 of 17 papers
- Ref-COCO · 2 of 17 papers
- Ref-COCOg · 2 of 17 papers
- OVIS · 2 of 17 papers
Baselines
- XMem · 6 of 17 papers
- STCN · 5 of 17 papers
- DEVA · 4 of 17 papers
- SAM 2.1 · 3 of 17 papers
- Cutie · 3 of 17 papers
- LMPM · 2 of 17 papers
- DsHmp · 2 of 17 papers
- MTTR · 2 of 17 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 57 papers; from OpenAlex and the papers' affiliations).
- Fudan University · 3 papers
- Peking University · 2 papers
- University of Science and Technology of China · 2 papers
- Microsoft · 2 papers
- Hangzhou Dianzi University · 2 papers
- Zhejiang University · 2 papers
- Shanghai AI Laboratory · 2 papers
- Nanyang Technological University · 2 papers
- Tongji University · 2 papers
- The Hong Kong Polytechnic University · 2 papers
Use it
Download BibTeX of 38 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
- Mask propagation for unsupervised video segmentation (67 papers)
- Video semantic and panoptic segmentation (67 papers)
Elsewhere in the topic
- Self-supervised video representation learning (167 papers)
- Video restoration with temporal priors (159 papers)
- Layered video decomposition and dynamics (140 papers)
- Stochastic video prediction (125 papers)
- Physics-inspired motion prediction (113 papers)
- Sign language and physiological video analysis (99 papers)
- Weakly supervised video anomaly detection (99 papers)
- Spatio-temporal predictive modeling (91 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).