Research map / Video / Action recognition
End-To-End Learning of Action Detection From Frame Glimpses in Videos
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use recurrent observation and direct temporal-bound prediction instead of sliding-window detection | Action detection · THUMOS'14 | mAP @ IoU 0.5: 14.4 -> 17.1 | Helped |
| Use recurrent observation and direct temporal-bound prediction instead of sliding-window detection | Action detection · ActivityNet Sports validation | mAP @ IoU 0.5: 33.2 -> 36.7 | Helped |
| Use recurrent observation and direct temporal-bound prediction instead of sliding-window detection | Action detection · ActivityNet Work validation | mAP @ IoU 0.5: 31.1 -> 39.9 | Helped |
| Add a learned prediction indicator that chooses when to emit candidate detections (ablation) | End-to-end action detector · THUMOS'14 | mAP @ IoU 0.5: 12.4 -> 17.1 | Helped |
| Add a learned observation-location policy rather than uniformly sampling frames (ablation) | End-to-end action detector · THUMOS'14 | mAP @ IoU 0.5: 9.3 -> 17.1 | Helped |
| Add both the learned observation-location policy and prediction indicator (ablation) | End-to-end action detector · THUMOS'14 | mAP @ IoU 0.5: 8.6 -> 17.1 | Helped |
| Add temporal localization regression (ablation) | End-to-end action detector · THUMOS'14 | mAP @ IoU 0.5: 5.5 -> 17.1 | Helped |
| Replace direct temporal-bound prediction with dense CNN frame probabilities and NMS (ablation) | Action detector · THUMOS'14 | mAP @ IoU 0.5: 17.1 -> 6.4 | Hurt |
| Add LSTM temporal smoothing to frame-level probabilities before NMS (ablation) | CNN frame-level detector · THUMOS'14 | mAP @ IoU 0.5: 6.4 -> 5.6 | Hurt |
| Vary the number of observations per video sequence among 4, 6, 8, and 10 (ablation) | End-to-end action detector · THUMOS'14 | Detection performance | No clear effect |
| Use learned sparse frame observations rather than exhaustive frame processing | End-to-end action detector · THUMOS'14 | Fraction of video frames observed: 2% or less | Helped |
Closest papers by meaning
- An Empirical Study of End-to-End Temporal Action Detection (CVPR 2022 · similarity 0.90)
- Action-Guided Attention for Video Action Anticipation (ICLR 2026 · similarity 0.89)
- SAP: Self-Adaptive Proposal Model for Temporal Action Detection Based on Reinforcement Learning (AAAI 2018 · similarity 0.89)
- Learning to Discriminate Information for Online Action Detection (CVPR 2020 · similarity 0.88)
- Three Birds with One Stone: Multi-Task Temporal Action Detection via Recycling Temporal Annotations (CVPR 2021 · similarity 0.88)
- Context-Enhanced Memory-Refined Transformer for Online Action Detection (CVPR 2025 · similarity 0.87)
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.