Research map / Video / Action recognition
ActionFormer: Localizing Moments of Actions with Transformers
Chen-Lin Zhang, Jianxin Wu, Yin Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace convolutional backbone with Transformer (ablation) | AF Base · THUMOS14 | average mAP: 46.6 -> 62.7 | Improved |
| add layer norm in heads (ablation) | ActionFormer · THUMOS14 | average mAP: 62.7 -> 65.4 | Improved |
| add center sampling (ablation) | ActionFormer · THUMOS14 | average mAP: 65.4 -> 66.8 | Improved |
| add position encoding (ablation) | ActionFormer · THUMOS14 | average mAP: 66.8 -> 66.7 | No clear effect |
| add score fusion (ablation) | ActionFormer · THUMOS14 | average mAP: 66.8 -> 62.1 | Hurt |
| add multiscale feature pyramid (vs single level) (ablation) | ActionFormer · THUMOS14 | average mAP: 47.6 -> 66.8 | Improved |
| use local window 19 instead of global attention (ablation) | ActionFormer · THUMOS14 | average mAP: 66.8 -> 66.8 | No clear effect |
| initial regression range [0,4) vs [0,16) (ablation) | ActionFormer (3 levels) · THUMOS14 | average mAP: 50.2 -> 60.1 | Improved |
| increase pyramid levels from 3 to 6 (ablation) | ActionFormer · THUMOS14 | average mAP: 60.1 -> 66.8 | Improved |
| increase pyramid levels from 6 to 7 (ablation) | ActionFormer · THUMOS14 | average mAP: 66.8 -> 66.2 | No clear effect |
| vary regression loss weight lambda_reg from 1 (ablation) | ActionFormer · THUMOS14 | average mAP: 65.4 to 66.8 | No clear effect |
| increase max training length from 576 to 2304 (ablation) | ActionFormer · THUMOS14 | average mAP: 65.7 -> 66.8 | Improved |
Later papers that compared with this method
21 later accepted papers report ActionFormer as a baseline in their results tables; 21 of them report a better number on most of the results they share with it.
- ActivityForensics: A Comprehensive Benchmark for Localizing Manipulated Activity in Videos (CVPR 2026 · better on 10 of 10 shared results)
- SkillSpotter: Pose-Aware Multi-View Skilled Action Detection and Grading in Ego-Exo Videos (ECCV 2026 · better on 7 of 7 shared results)
- MG-RWKV: Multi-Grained Context-Aware RWKV for Temporal Forgery Localization (ECCV 2026 · better on 4 of 4 shared results)
- HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding (NeurIPS 2026 · better on 2 of 2 shared results)
- SAVA-X: Ego-to-Exo Imitation Error Detection via Scene-Adaptive View Alignment and Bidirectional Cross View Fusion (CVPR 2026 · better on 2 of 2 shared results)
- Efficient Action Counting with Dynamic Queries (IJCV 2026 · better on 2 of 2 shared results)
- DeFB: Decomposed Feature Learning for Real-Time Multi-Person Eyeblink Detection in Untrimmed In-the-Wild Videos (AAAI 2026 · better on 1 of 1 shared result)
- Energy vs. Noise: Towards Robust Temporal Action Localization in Open-World (AAAI 2025 · better on 4 of 4 shared results)
13 more not listed.
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- G-TAD — G-TAD: Sub-Graph Localization for Temporal Action Detection (CVPR 2020) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TALLFormer: Temporal Action Localization with a Long-Memory Transformer (ECCV 2022 · similarity 0.92)
- Graph Convolutional Module for Temporal Action Localization in Videos (TPAMI 2021 · similarity 0.91)
- Video Action Transformer Network (CVPR 2019 · similarity 0.91)
- An Efficient Spatio-Temporal Pyramid Transformer for Action Detection (ECCV 2022 · similarity 0.90)
- Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer (CVPR 2025 · similarity 0.88)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
