Research map / Video / Action recognition
Leaving Some Stones Unturned: Dynamic Feature Prioritization for Activity Detection in Streaming Video
Yu-Chuan Su, Kristen 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 |
|---|---|---|---|
| learn dynamic feature-prioritization policy instead of random (Passive) selection | Streaming activity recognition · UCF-101 and ADL streaming | cost to reach accuracy: reduces cost by 80% and 50% on UCF and ADL | Improved |
| use instance-dependent feature selection instead of DT-Top static top features | Streaming recognition · UCF-101 multi-class | accuracy | Improved |
| add skip action and dynamic policy with CNN features | Streaming recognition with CNN · UCF-101 | frames processed: skips 80% of the frames, achieves over 90% of the ultimate accuracy | Improved |
| dynamic policy instead of Passive with CNN features | Streaming recognition with CNN · UCF-101 | frames needed: less than half of the frames | Improved |
| dynamic policy for batch recognition | Batch recognition · ADL | features extracted for same accuracy: 30% of the features | Improved |
| dynamic policy instead of Object-Preference | Batch recognition · UCF-101 and ADL | speed: 2.5 times faster | Improved |
| dynamic policy vs DT-Top in untrimmed binary detection | Untrimmed streaming detection · UCF-101 | F1 | No clear effect |
| dynamic policy vs baselines in untrimmed detection | Untrimmed streaming detection · ADL | F1: nearly twice the accuracy of all baselines until 16 fps | Improved |
| dynamic policy with skip action at high detector speed | Untrimmed streaming detection · UCF-101 at 64 fps | detections run: 50% fewer detections | Improved |
| dynamic policy instead of MMED early detector | Untrimmed streaming detection · ADL and UCF AMOC | timeliness and cost | Improved |
| train decision tree on same features as classifier f without subvolumes (ablation) | Decision tree baseline · Batch recognition | accuracy | Hurt |
| add dense trajectories to CNN features | Streaming recognition · UCF-101 | accuracy | Improved |
Closest papers by meaning
- HARDVS: Revisiting Human Activity Recognition with Dynamic Vision Sensors (AAAI 2024 · similarity 0.88)
- Discriminative Hierarchical Rank Pooling for Activity Recognition (CVPR 2016 · similarity 0.87)
- Two-Stream Region Convolutional 3D Network for Temporal Activity Detection (TPAMI 2019 · similarity 0.87)
- Fast Temporal Activity Proposals for Efficient Detection of Human Actions in Untrimmed Videos (CVPR 2016 · similarity 0.87)
- Efficient Activity Detection in Untrimmed Video with Max-Subgraph Search (TPAMI 2016 · similarity 0.87)
- Weakly-Guided Self-Supervised Pretraining for Temporal Activity Detection (AAAI 2023 · 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.
