Research map / Video / Action recognition
Action Search: Spotting Actions in Videos and Its Application to Temporal Action Localization
Humam Alwassel, Fabian Caba Heilbron, Bernard Ghanem
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 learned Action Search as precursor to Res3D + S-CNN | Res3D + S-CNN · THUMOS14 test, tIoU 0.5 | mAP: 22.5 -> 30.8 | Improved |
| add learned Action Search as precursor to Res3D + S-CNN | Res3D + S-CNN · THUMOS14 test | observed frames S: 100 -> 17.3 | Improved |
| train search by imitating human searches instead of reinforcement policy | Frame Glimpses · THUMOS14 test, tIoU 0.5 | mAP: 17.1 -> 30.8 | Improved |
| use ResNet backbone instead of C3D (ablation) | Action Search · THUMOS14 test, tIoU 0.5 | mAP: 23.9 -> 30.8 | Improved |
| use ResNet backbone instead of C3D (ablation) | Action Search · THUMOS14 test | observed frames S: 45.1 -> 17.3 | Improved |
| add Action Search with C3D to the C3D-based detector (ablation) | Res3D + S-CNN baseline · THUMOS14 test, tIoU 0.5 | mAP: +4.9% mAP, 54.9% less frames observed | Improved |
| train on 25% of human searches (ablation) | Action Search · THUMOS14 test, tIoU 0.5 | mAP: 22.5 -> 22.4 | No clear effect |
| train on 25% of human searches (ablation) | Action Search · THUMOS14 test | observed frames S: 100 -> 63.5 | Improved |
| train on 50% instead of 25% of human searches (ablation) | Action Search · THUMOS14 test, tIoU 0.5 | mAP: 22.4 -> 26.1 | Improved |
| train on 100% instead of 50% of human searches (ablation) | Action Search · THUMOS14 test, tIoU 0.5 | mAP: 26.1 -> 30.8 | Improved |
| train on 100% instead of 50% of human searches (ablation) | Action Search · THUMOS14 test | observed frames S: 34.4 -> 17.3 | Improved |
| add Action Search with temporal context vs direction-only baseline | Direction Baseline · AVA test (coverage <=5%) | observations to spot action: 16.6% fewer observations | Improved |
Closest papers by meaning
- Graph Convolutional Module for Temporal Action Localization in Videos (TPAMI 2021 · similarity 0.92)
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.92)
- Learning To Refactor Action and Co-Occurrence Features for Temporal Action Localization (CVPR 2022 · similarity 0.92)
- One-Shot Action Localization by Learning Sequence Matching Network (CVPR 2018 · similarity 0.91)
- AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions (CVPR 2018 · similarity 0.91)
- Diagnosing Error in Temporal Action Detectors (ECCV 2018 · similarity 0.91)
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.
