Research map / Video / Action recognition
Weakly supervised temporal action localization
Localizes actions using video-level labels with snippet modeling, uncertainty and graph networks.
79 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Action recognition, keyword cluster “temporal action · action localization · action detection”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
16 papers in 2023–24 and 11 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 11.3% to 8.3%.
No longer reported by papers here: THUMOS14 test (mAP@0.5, last in 2023); THUMOS14 (mAP@0.5, last in 2024); THUMOS14 test (mAP@IoU=0.5, last in 2023); ActivityNet v1.3 validation (average mAP, last in 2024); THUMOS14 (mAP@IoU 0.5, last in 2023).
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 · Temporal Action Localization in Untrimmed Videos via Multi-Stage CNNs (CVPR 2016 · 967 citations)
- 2017 · CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · 608 citations)
- 2018 · Rethinking the Faster R-CNN Architecture for Temporal Action Localization (CVPR 2018 · 781 citations)
- 2019 · Gaussian Temporal Awareness Networks for Action Localization (CVPR 2019 · 390 citations)
- 2020 · Background Suppression Network for Weakly-Supervised Temporal Action Localization (AAAI 2020 · 262 citations)
- 2021 · Learning Salient Boundary Feature for Anchor-free Temporal Action Localization (CVPR 2021 · 288 citations)
- 2022 · ASM-Loc: Action-Aware Segment Modeling for Weakly-Supervised Temporal Action Localization (CVPR 2022 · 114 citations)
- 2023 · Proposal-Based Multiple Instance Learning for Weakly-Supervised Temporal Action Localization (CVPR 2023 · 61 citations)
- 2024 · End-to-End Spatio-Temporal Action Localisation with Video Transformers (CVPR 2024 · 30 citations)
- 2025 · Weakly Supervised Temporal Action Localization via Dual-Prior Collaborative Learning Guided by Multimodal Large Language Models (CVPR 2025 · 16 citations)
- 2026 · What-Meets-Where: Unified Learning of Action and Contact Localization in Images (AAAI 2026 · 2 citations)
Earliest papers here
- Temporal Action Localization in Untrimmed Videos via Multi-Stage CNNs (CVPR 2016 · 967 citations)
- Temporal Action Localization With Pyramid of Score Distribution Features (CVPR 2016 · 214 citations)
- Predicting the Where and What of Actors and Actions Through Online Action Localization (CVPR 2016 · 93 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.
THUMOS14 test · mAP@0.5 (higher is better)
6 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 38.8 | Gaussian Temporal Awareness Networks for Action Localization (CVPR 2019 · 390 citations) |
| 2022 | 52.9 | RCL: Recurrent Continuous Localization for Temporal Action Detection (CVPR 2022 · 57 citations) |
THUMOS14 · mAP@0.5 (higher is better)
5 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 28.8 | Weakly-Supervised Action Localization by Generative Attention Modeling (CVPR 2020 · 183 citations) |
| 2021 | 55.5 | Learning Salient Boundary Feature for Anchor-free Temporal Action Localization (CVPR 2021 · 288 citations) |
| 2024 | 64.26 | Boundary Denoising for Video Activity Localization (ICLR 2024 · 5 citations) |
THUMOS14 test · mAP@IoU=0.5 (higher is better)
5 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 29 | ActionBytes: Learning From Trimmed Videos to Localize Actions (CVPR 2020 · 62 citations) |
| 2021 | 35.9 | Uncertainty Guided Collaborative Training for Weakly Supervised Temporal Action Detection (CVPR 2021 · 109 citations) |
| 2023 | 39.3 | Boosting Weakly-Supervised Temporal Action Localization With Text Information (CVPR 2023 · 32 citations) |
ActivityNet v1.3 validation · average mAP (higher is better)
4 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 34.31 | Gaussian Temporal Awareness Networks for Action Localization (CVPR 2019 · 390 citations) |
| 2022 | 38.6 | Learning To Refactor Action and Co-Occurrence Features for Temporal Action Localization (CVPR 2022 · 47 citations) |
THUMOS14 · mAP@IoU 0.5 (higher is better)
4 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 27 | Background Suppression Network for Weakly-Supervised Temporal Action Localization (AAAI 2020 · 262 citations) |
| 2021 | 56.9 | Multi-Shot Temporal Event Localization: A Benchmark (CVPR 2021 · 79 citations) |
What has been tried
73 papers here report 777 measured changes (a component added, removed or swapped, compared with a baseline): helped 80%, no clear effect 3%, hurt 15%, 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 |
|---|---|---|---|---|
| Bind each query to a person rather than to an action | AVA · mAP | 23.6 -> 26.7; 25.7 -> 27.8 | Helped | End-to-End Spatio-Temporal Action Localisation with Video Transformers (CVPR 2024 · 30 citations) |
| Add L_contra to the baseline objective | THUMOS14 · mAP@IoU [0.1:0.7] | 49.4 -> 51.5 | Helped | HR-Pro: Point-Supervised Temporal Action Localization via Hierarchical Reliability Propagation (AAAI 2024 · 28 citations) |
| Use L1 distance instead of random salient assignment | THUMOS14 · average mAP | 42.3 -> 46.8 | Helped | Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · 26 citations) |
| Add proposal learning to the baseline | THUMOS’14 · mAP@IoU (0.1:0.7) | 52.5 -> 55.0 | Helped | Realigning Confidence with Temporal Saliency Information for Point-Level Weakly-Supervised Temporal Action Localization (CVPR 2024 · 17 citations) |
| Add KSM | THUMOS14 · average mAP@IoU | 37.4 -> 38.9 | Helped | Weakly Supervised Temporal Action Localization via Dual-Prior Collaborative Learning Guided by Multimodal Large Language Models (CVPR 2025 · 16 citations) |
| Add representation loss to BCE objective | THUMOS14 50%-50% · average mAP | 7.9 -> 10.3 | Helped | Test-Time Zero-Shot Temporal Action Localization (CVPR 2024 · 15 citations) |
| add boundary-denoising training | QVHighlights val · mAP Avg. | 44.56 vs 42.95 without denoise | Helped | Boundary Denoising for Video Activity Localization (ICLR 2024 · 5 citations) |
| add the SME loss | THUMOS14 · AVG mAP | 43.1 -> 50.9 | Helped | Similar Modality Enhancement and Action Consistency Learning for Weakly Supervised Temporal Action Localization (AAAI 2025 · 4 citations) |
| Add GMG instant, local, and global granularity branches | THUMOS14 · average mAP | 73.9 -> 74.3 | Helped | Temporal Action Localization with Cross Layer Task Decoupling and Refinement (AAAI 2025 · 4 citations) |
| Add Meta-learning Pipeline | THUMOS14, mixed noise 0.5×50% · average mAP (%) | 44.26 -> 50.05 | Helped | Energy vs. Noise: Towards Robust Temporal Action Localization in Open-World (AAAI 2025 · 2 citations) |
| Replace extended point-label pseudo segments with optimal-transport pseudo labels | THUMOS14 · mAP@[0.3:0.7] | 2.2 -> 22.9 | Helped | Boosting Point-Supervised Temporal Action Localization through Integrating Query Reformation and Optimal Transport (CVPR 2025 · 2 citations) |
| Add the global loss to the local loss | GroundingYouTube · mAP IoU@0.4 | 7.29 -> 19.45 | Helped | What When and Where? Self-Supervised Spatio-Temporal Grounding in Untrimmed Multi-Action Videos from Narrated Instructions (CVPR 2024 · 2 citations) |
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
- THUMOS14 · 5 of 6 papers
- ActivityNet1.3 · 3 of 6 papers
Baselines
- ActionFormer · 3 of 6 papers
- TriDet · 3 of 6 papers
- TemporalMaxer · 2 of 6 papers
- UM · 2 of 6 papers
- CoLA · 2 of 6 papers
- CO2-Net · 2 of 6 papers
- DELU · 2 of 6 papers
- PivoTAL · 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 27 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 3 papers
- Xidian University · 2 papers
- Shanghai Jiao Tong University · 2 papers
- King Abdullah University of Science and Technology · 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
- Temporal action segmentation (88 papers)
- Fully supervised temporal action detection (87 papers)
- Action anticipation and latent actions (79 papers)
Elsewhere in the topic
- Disentangled action recognition (182 papers)
- Spatiotemporal CNNs for action recognition (82 papers)
- Contrastive video representation learning (74 papers)
- Skeleton-based action recognition (46 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).