Research map / Video / Action recognition
GateHUB: Gated History Unit With Background Suppression for Online Action Detection
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 GateHUB instead of LSTR with TSN RGB and optical flow features | Online action detection · THUMOS'14 | mAP (%): 69.5 -> 70.7 | Helped |
| Use GateHUB instead of LSTR with TimeSformer RGB and TSN optical flow features | Online action detection · THUMOS'14 | mAP (%): 69.6 -> 72.5 | Helped |
| Use GateHUB instead of LSTR | Online action detection · TVSeries | mcAP (%): 89.1 -> 89.6 | Helped |
| Use GateHUB instead of OadTR | Online action detection · HDD | mAP (%): 29.8 -> 32.1 | Helped |
| Add GHU instead of vanilla cross-attention (ablation) | GateHUB · THUMOS'14 | mAP (%): 69.6 -> 70.7 | Helped |
| Use both enhancement and suppression in GHU rather than either operation alone (ablation) | GateHUB · THUMOS'14 | mAP (%): 70.5 -> 70.7 | Helped |
| Add position guidance to GHU (ablation) | GateHUB · THUMOS'14 | mAP (%): 70.3 -> 70.7 | Helped |
| Share GHU gating scores across heads rather than learning them per head (ablation) | GateHUB · THUMOS'14 | mAP (%): 68.0 -> 70.7 | Helped |
| Use separate action and background emphasis with gamma_a > gamma_b instead of cross-entropy or standard focal loss (ablation) | GateHUB · THUMOS'14 | mAP (%): 69.9 -> 70.7; 70.2 -> 70.7 | Helped |
| Set gamma_a < gamma_b rather than gamma_a > gamma_b (ablation) | GateHUB background suppression objective · THUMOS'14 | mAP (%): 70.7 -> 70.2 | Hurt |
| Add FaH with 2s of future history (ablation) | GateHUB · THUMOS'14 | mAP (%): 71.5 -> 72.5 | Helped |
| Use FaH with 0.5s or 4s of future history instead of 2s (ablation) | GateHUB · THUMOS'14 | mAP (%): 72.5 -> 71.1; 72.5 -> 71.4 | Hurt |
Closest papers by meaning
- Context-Enhanced Memory-Refined Transformer for Online Action Detection (CVPR 2025 · similarity 0.86)
- History-Guided Video Diffusion (ICML 2025 · similarity 0.84)
- Energy vs. Noise: Towards Robust Temporal Action Localization in Open-World (AAAI 2025 · similarity 0.84)
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (CVPR 2024 · similarity 0.84)
- ActFusion: a Unified Diffusion Model for Action Segmentation and Anticipation (NeurIPS 2024 · similarity 0.83)
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.83)
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.