Research map / Video / Action recognition
Learning Actionness via Long-range Temporal Order Verification
Dimitri Zhukov, Jean-Baptiste Alayrac, Ivan Laptev, Josef Sivic
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 |
|---|---|---|---|
| add actionness score to inference of Zhukov et al. step localization | Zhukov et al. [41] · CrossTask step localization | average recall (%): 22.4 -> 27.0 | Improved |
| add actionness score to inference of Miech et al. step localization | Miech et al. [23] · CrossTask step localization | average recall (%): 33.6 -> 34.0 | Improved |
| add actionness score to I3D HowTo100M step localization | I3D [22] · CrossTask step localization | average recall (%): 36.4 -> 37.6 | Improved |
| add actionness score to S3D HowTo100M step localization | S3D [22] · CrossTask step localization | average recall (%): 40.3 -> 41.0 | Improved |
| trim 30% of frames at start and end of training videos (ablation) | order-verification actionness model · COIN and CrossTask action vs background | average precision | Improved |
| use long-range (minutes) instead of short-range (10-20 seconds) order verification (ablation) | order-verification actionness model · COIN and CrossTask action vs background | average precision | Improved |
| increase number of sampled segments M above 5 (ablation) | order-verification actionness model · COIN and CrossTask action vs background | average precision | Hurt |
| sample only 3 segments instead of 5 per video (ablation) | order-verification actionness model · COIN and CrossTask action vs background | average precision | Hurt |
| use learned actionness instead of chance with trimming | actionness score · COIN action vs background | average precision (%): 53.5% -> 59.0% | Improved |
| use learned actionness instead of chance with trimming | actionness score · CrossTask action vs background | average precision (%): 32.6% -> 47.6% | Improved |
| add hand detector or optical flow actionness instead of chance | hand detector / optical flow · COIN action vs background | average precision (%): 45.6% -> 50.6% | No clear effect |
Closest papers by meaning
- Completeness Modeling and Context Separation for Weakly Supervised Temporal Action Localization (CVPR 2019 · similarity 0.92)
- Weakly-Supervised Action Localization With Background Modeling (ICCV 2019 · similarity 0.91)
- Learning Action Completeness From Points for Weakly-Supervised Temporal Action Localization (ICCV 2021 · similarity 0.91)
- Temporal Action Localization by Structured Maximal Sums (CVPR 2017 · similarity 0.91)
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.91)
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.90)
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.
