Research map / Video / Action recognition
Efficient Activity Detection in Untrimmed Video with Max-Subgraph Search
Chao-Yeh Chen, Kristen Lorraine Grauman
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 temporal jumps (T-Jump) to T-Subgraph (ablation) | T-Subgraph · UCF Sports | mean overlap accuracy: Diving 0.8106 -> 0.9091; Riding 0.5349 -> 0.3888 | Mixed |
| add temporal jumps (T-Jump) to T-Subgraph (ablation) | T-Subgraph · Hollywood uncropped | mean overlap accuracy: AnswerPhone 0.3968 -> 0.3994; GetOutCar 0.2276 -> 0.2921 | Improved |
| use T-Subgraph search instead of exhaustive T-Sliding | T-Subgraph · UCF Sports | detection time (ms): 1.25x10^5 -> 1.02x10^2 | Improved |
| use T-Subgraph search instead of exhaustive T-Sliding | T-Subgraph · Hollywood uncropped | detection time (ms): 3.71x10^3 -> 6.63x10 | Improved |
| use T-Subgraph instead of branch-and-bound ST-Cube-Subvolume | T-Subgraph · Hollywood uncropped | detection time (ms): 1.70x10^5 -> 6.63x10 | Improved |
| add temporal jumps (T-Jump), more complex graph (ablation) | T-Subgraph · Hollywood uncropped | detection time (ms): 6.63x10 -> 5.69x10^2 | Hurt |
| detect activity first (T-Subgraph output) before recognition instead of raw uncropped clips | recognition pipeline · Hollywood | recognition accuracy: 24.83% -> 29.66% | Improved |
| replace HoG/HoF low-level features with object-based high-level descriptors (ablation) | T-Subgraph · Hollywood uncropped | mean overlap accuracy: HugPerson 0.3869 -> 0.5292; AnswerPhone 0.3968 -> 0.1741 | Mixed |
| use T-Subgraph vs T-Sliding for multiple instances | T-Subgraph · THUMOS 2014 | mAP: 0.1983 -> 0.2143 | Improved |
| add temporal jumps (T-Jump) to T-Subgraph (ablation) | T-Subgraph · THUMOS 2014 | mAP: 0.2143 -> 0.1546 | Hurt |
| add temporal jumps (T-Jump) to T-Subgraph (ablation) | T-Subgraph · THUMOS 2014 | overlap: 0.2186 -> 0.2636 | Improved |
| add spatial links (ST-Subgraph) to temporal subgraph (ablation) | T-Subgraph · MSR Action temporal overlap | mean temporal overlap accuracy: Boxing 0.0541 -> 0.0989; Clapping 0.0982 -> 0.1754 | Improved |
Closest papers by meaning
- Fast Temporal Activity Proposals for Efficient Detection of Human Actions in Untrimmed Videos (CVPR 2016 · similarity 0.89)
- R-C3D: Region Convolutional 3D Network for Temporal Activity Detection (ICCV 2017 · similarity 0.89)
- Deep Learning-Based Action Detection in Untrimmed Videos: A Survey (TPAMI 2022 · similarity 0.88)
- Sum Product Networks for Activity Recognition (TPAMI 2016 · similarity 0.88)
- Two-Stream Region Convolutional 3D Network for Temporal Activity Detection (TPAMI 2019 · similarity 0.88)
- Boundary Denoising for Video Activity Localization (ICLR 2024 · similarity 0.86)
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.
