Research map / Video / Action recognition
CAG-QIL: Context-Aware Actionness Grouping via Q Imitation Learning for Online Temporal Action Localization
Hyolim Kang, Kyungmin Kim, Yumin Ko, Seon Joo Kim
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 class probability p to the agent state (ablation) | CAG-QIL · THUMOS14 On-TAL tIoU 0.5 | mAP: 27.2 -> 29.8 | Improved |
| add class probability p to the agent state (ablation) | CAG-QIL · THUMOS14 On-TAL tIoU 0.3 | mAP: 43.0 -> 44.7 | Improved |
| add class probability p to the agent state (ablation) | CAG-QIL · THUMOS14 On-TAL tIoU 0.7 | mAP: 12.4 -> 14.5 | Improved |
| train the agent with Q imitation learning instead of hand-crafted reward RL (ablation) | CAG · THUMOS14 On-TAL tIoU 0.5 | mAP: 22.2 -> 29.8 | Improved |
| train the agent with Q imitation learning instead of behavior cloning (ablation) | CAG · THUMOS14 On-TAL tIoU 0.5 | mAP: 3.3 -> 29.8 | Improved |
| train with QIL instead of RL (ablation) | CAG · ActivityNet1.3 val tIoU 0.5 | mAP: 21.4 -> 30.5 | Improved |
| train with QIL instead of BC (ablation) | CAG · ActivityNet1.3 val tIoU 0.5 | mAP: 9.5 -> 30.5 | Improved |
| add the CAG-QIL context-aware agent on top of OAD grouping | OAD-Grouping · THUMOS14 On-TAL tIoU 0.5 | mAP: 22.0 -> 29.8 | Improved |
| add the CAG-QIL context-aware agent on top of OAD grouping | OAD-Grouping · ActivityNet1.3 val tIoU 0.75 | mAP: 15.7 -> 18.5 | Improved |
| train the agent with hand-crafted frame-wise reward RL | OAD-Grouping · THUMOS14 On-TAL (4 trials) | mAP difference to baseline: 0.4 (SD 1.5) at tIoU 0.3, -0.1 (SD 0.8) at 0.5 | No clear effect |
| train the agent with behavior cloning | OAD-Grouping · THUMOS14 On-TAL (4 trials) | mAP difference to baseline: -25.2 at tIoU 0.3, -17.2 at 0.5 | Hurt |
| use CAG-QIL instead of hindsight per-class threshold | OAD-Grouping · THUMOS14 On-TAL tIoU 0.5 | mAP: 24.5 -> 29.8 | Improved |
Later papers that compared with this method
3 later accepted papers report CAG-QIL as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- ActionSwitch: Class-agnostic Detection of Simultaneous Actions in Streaming Videos (ECCV 2024 · better on 4 of 4 shared results)
- Online Temporal Action Localization with Memory-Augmented Transformer (ECCV 2024 · better on 1 of 1 shared result)
- A Sliding Window Scheme for Online Temporal Action Localization (ECCV 2022 · better on 11 of 11 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
- Action Sensitivity Learning for Temporal Action Localization (ICCV 2023 · similarity 0.89)
- PivoTAL: Prior-Driven Supervision for Weakly-Supervised Temporal Action Localization (CVPR 2023 · similarity 0.89)
- ActionSwitch: Class-agnostic Detection of Simultaneous Actions in Streaming Videos (ECCV 2024 · similarity 0.89)
- Graph Convolutional Module for Temporal Action Localization in Videos (TPAMI 2021 · similarity 0.89)
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.88)
- Learning To Refactor Action and Co-Occurrence Features for Temporal Action Localization (CVPR 2022 · 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.
