Research map / Video / Video segmentation and tracking
Sign language and physiological video analysis
Applies temporal modeling to sign language recognition, visual speech and remote physiological measurement.
99 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Video segmentation and tracking, keyword cluster “detection · anomaly · supervised”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 4 of 4 tracked benchmarks.
29 papers in 2023–24 and 34 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 6.6% to 6.3%.
Open questions the data points to
- CSL-Daily Test (WER): best went from 29.3 (up to 2024) to 24.7 in 2025 — OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition (AAAI 2025 · 6 citations)
- CSL-Daily Dev (WER): best went from 29.7 (up to 2024) to 25.8 in 2025 — OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition (AAAI 2025 · 6 citations)
- PHOENIX14 Test (WER): best went from 18.9 (up to 2024) to 17.2 in 2025 — OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition (AAAI 2025 · 6 citations)
- PHOENIX14-T Test (WER): best went from 19.4 (up to 2024) to 18.4 in 2025 — OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition (AAAI 2025 · 6 citations)
- Papers report 1-Mio-Hands on RWTH-PHOENIX-Weather 2014 test (WER) anywhere from 45.1 to 50.2: comparisons against it depend on whose number you take.
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 · Multimodal Spontaneous Emotion Corpus for Human Behavior Analysis (CVPR 2016 · 437 citations)
- 2017 · Recurrent Convolutional Neural Networks for Continuous Sign Language Recognition by Staged Optimization (CVPR 2017 · 350 citations)
- 2018 · Neural Sign Language Translation (CVPR 2018 · 787 citations)
- 2019 · Iterative Alignment Network for Continuous Sign Language Recognition (CVPR 2019 · 282 citations)
- 2020 · Spatial-Temporal Multi-Cue Network for Continuous Sign Language Recognition (AAAI 2020 · 222 citations)
- 2021 · Positive Sample Propagation Along the Audio-Visual Event Line (CVPR 2021 · 118 citations)
- 2022 · PhysFormer: Facial Video-Based Physiological Measurement With Temporal Difference Transformer (CVPR 2022 · 293 citations)
- 2023 · Continuous Sign Language Recognition With Correlation Network (CVPR 2023 · 145 citations)
- 2024 · LoCoNet: Long-Short Context Network for Active Speaker Detection (CVPR 2024 · 28 citations)
- 2025 · Discrete to Continuous: Generating Smooth Transition Poses from Sign Language Observations (CVPR 2025 · 18 citations)
- 2026 · EVOKE: Efficient and High-Fidelity EEG-to-Video Reconstruction via Decoupling Implicit Neural Representation (AAAI 2026 · 1 citation)
Earliest papers here
- Multimodal Spontaneous Emotion Corpus for Human Behavior Analysis (CVPR 2016 · 437 citations)
- Automating Carotid Intima-Media Thickness Video Interpretation With Convolutional Neural Networks (CVPR 2016 · 60 citations)
- Recurrent Convolutional Neural Networks for Continuous Sign Language Recognition by Staged Optimization (CVPR 2017 · 350 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.
CSL-Daily Test · WER (lower is better)
5 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 29.4 | Distilling Cross-Temporal Contexts for Continuous Sign Language Recognition (CVPR 2023 · 59 citations) |
| 2024 | 29.3 | TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · 23 citations) |
| 2025 | 24.7 | OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition (AAAI 2025 · 6 citations) |
CSL-Daily Dev · WER (lower is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 30.6 | Continuous Sign Language Recognition With Correlation Network (CVPR 2023 · 145 citations) |
| 2024 | 29.7 | TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · 23 citations) |
| 2025 | 25.8 | OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition (AAAI 2025 · 6 citations) |
PHOENIX14 Test · WER (lower is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 19.4 | Continuous Sign Language Recognition With Correlation Network (CVPR 2023 · 145 citations) |
| 2024 | 18.9 | TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · 23 citations) |
| 2025 | 17.2 | OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition (AAAI 2025 · 6 citations) |
PHOENIX14-T Test · WER (lower is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 20.5 | Continuous Sign Language Recognition With Correlation Network (CVPR 2023 · 145 citations) |
| 2024 | 19.4 | TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · 23 citations) |
| 2025 | 18.4 | OLMD: Orientation-aware Long-term Motion Decoupling for Continuous Sign Language Recognition (AAAI 2025 · 6 citations) |
What has been tried
77 papers here report 828 measured changes (a component added, removed or swapped, compared with a baseline): helped 75%, no clear effect 4%, hurt 19%, 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 |
|---|---|---|---|---|
| Replace ResNet-34 with VGGFrame as the audio encoder | AVA-ActiveSpeaker · mAP | 91.2 -> 92.8 | Helped | LoCoNet: Long-Short Context Network for Active Speaker Detection (CVPR 2024 · 28 citations) |
| Add the trajectory module | PHOENIX14 · WER | 20.2 -> 19.2; 21.0 -> 19.4 | Helped | TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · 23 citations) |
| increase masking ratio to 95% | US video GBC detection · accuracy | 96.4% | Helped | FocusMAE: Gallbladder Cancer Detection from Ultrasound Videos with Focused Masked Autoencoders (CVPR 2024 · 20 citations) |
| Use masking ratio r=0.1 instead of r=0.3 | PHOENIX14T TEST, observation 20 frames - prediction 10 frames · DTW | 0.61 -> 0.60 | Helped | Discrete to Continuous: Generating Smooth Transition Poses from Sign Language Observations (CVPR 2025 · 18 citations) |
| Add contrastive loss | LRS2 · WER% | 29.87 -> 31.70 | Helped | Restoring Speaking Lips from Occlusion for Audio-Visual Speech Recognition (AAAI 2024 · 12 citations) |
| add Seq2Seq | 40-class SEED-DV · Video semantic accuracy, 2-way | 0.786±0.03 -> 0.798±0.03 | Helped | EEG2Video: Towards Decoding Dynamic Visual Perception from EEG Signals (NeurIPS 2024 · 12 citations) |
| Add interference disentanglement | MR-NIRP-DRV · RMSE | 8.12 -> 6.03 | Helped | Remote Photoplethysmography in Real-World and Extreme Lighting Scenarios (CVPR 2025 · 11 citations) |
| Use paired time-difference inputs for both RGB and RF instead of time-domain inputs | MMD-rPPG heart rate · MAE | 1.75 -> 1.61 | Helped | Fusion-Vital: Video-RF Fusion Transformer for Advanced Remote Physiological Measurement (AAAI 2024 · 11 citations) |
| Add the diffusion model | VIPL-HR · RMSE | 6.83 -> 6.37 | Helped | PhysDiff: Physiology-based Dynamicity Disentangled Diffusion Model for Remote Physiological Measurement (AAAI 2025 · 10 citations) |
| Add FSAM to FactorizePhys | UBFC-rPPG → PURE · MAE (HR) | 1.37 ± 1.02 -> 0.71 ± 0.39 | Helped | FactorizePhys: Matrix Factorization for Multidimensional Attention in Remote Physiological Sensing (NeurIPS 2024 · 9 citations) |
| Add spatial masking with ratio 0.2 | Validation set · SSIM | 0.171 -> 0.204 | Helped | NeuralFlix: A Simple While Effective Framework for Semantic Decoding of Videos from Non-invasive Brain Recordings (AAAI 2025 · 7 citations) |
| add GTT blocks to the baseline | EGait · accuracy | 79.8% -> 86.3% | Helped | GaitCycFormer: Leveraging Gait Cycles and Transformers for Gait Emotion Recognition. (AAAI 2025 · 6 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 |
|---|---|---|---|---|
| 1-Mio-Hands | RWTH-PHOENIX-Weather 2014 test | WER | 45.1 – 50.2 | 2 papers, e.g. CVPR 2019 and CVPR 2017 |
What reviewers will expect
Datasets and baselines named in the 19 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- PHOENIX14-T · 2 of 19 papers
- PURE · 2 of 19 papers
- UBFC-rPPG · 2 of 19 papers
- VIPL-HR · 2 of 19 papers
Baselines
- PhysFormer · 3 of 19 papers
- TS-CAN · 3 of 19 papers
- DeepPhys · 2 of 19 papers
- PhysNet · 2 of 19 papers
- NEST · 2 of 19 papers
- Dual-GAN · 2 of 19 papers
- EfficientPhys · 2 of 19 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 63 papers; from OpenAlex and the papers' affiliations).
- The Chinese University of Hong Kong · 4 papers
- Harbin Institute of Technology · 3 papers
- National University of Singapore · 3 papers
- Tianjin University · 2 papers
- Southern University of Science and Technology · 2 papers
- Samsung · 2 papers
- Google · 2 papers
- Beihang University · 2 papers
- Nanjing University · 2 papers
- Hefei University of Technology · 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
- Self-supervised video representation learning (167 papers)
- Video restoration with temporal priors (159 papers)
- Weakly supervised video anomaly detection (99 papers)
- Spatio-temporal predictive modeling (91 papers)
- Deepfake video detection (47 papers)
Elsewhere in the topic
- Layered video decomposition and dynamics (140 papers)
- Stochastic video prediction (125 papers)
- Physics-inspired motion prediction (113 papers)
- Matching-based video object segmentation (86 papers)
- Learned optical flow estimation (73 papers)
- Learned video compression (68 papers)
- Mask propagation for unsupervised video segmentation (67 papers)
- Video semantic and panoptic segmentation (67 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).