Research map / Video / Action recognition
Real-Time Online Video Detection with Temporal Smoothing Transformers
Yue Zhao, Philipp Krähenbühl
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 |
|---|---|---|---|
| use Laplace kernel (ES) instead of box kernel (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 16.14 -> 16.95 | Improved |
| exponential smoothing lambda=0 instead of FIFO for box kernel (ablation) | TeSTra with box kernel · EK100 anticipation (RGB) | unseen and tail action recall: 12.64 -> 12.22 | Hurt |
| stream-mode ES instead of batch-mode for Laplace kernel (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 16.95 -> 16.94 | No clear effect |
| add temporal PE to long-term memory (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 17.0 -> 16.8 | No clear effect |
| add temporal PE to short-term memory (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 15.7 -> 17.0 | Improved |
| add MixClip augmentation (rate 0.5) (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 15.5 -> 17.0 | Improved |
| MixClip rate 0.8 instead of 0.5 (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 17.0 -> 16.2 | Hurt |
| fuse short-term memory with compressed memory instead of no fuse (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 15.9 -> 17.0 | Improved |
| attach short-term memory to long memory instead of no fuse (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 15.9 -> 16.0 | No clear effect |
| add long memory without fusing short memory (ablation) | TeSTra · EK100 anticipation (RGB) | overall action recall: 16.1 -> 15.9 | Hurt |
| use TeSTra instead of LSTR (K400 features) | TeSTra · THUMOS'14 anticipation | average mAP: 52.6 -> 56.8 | Improved |
| NVOFA optical flow instead of TV-L1 | TeSTra · THUMOS'14 online detection | mAP: 71.2 -> 67.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OadTR — OadTR: Online Action Detection With Transformers (ICCV 2021) · better on 2 of 2 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
- Learning Streaming Video Representation via Multitask Training (ICCV 2025 · similarity 0.86)
- Time Is MattEr: Temporal Self-supervision for Video Transformers (ICML 2022 · similarity 0.86)
- E2E-LOAD++: End-to-end Long-form Online Action Detection and Beyond (TPAMI 2026 · similarity 0.86)
- Smooth Regularization for Efficient Video Recognition (NeurIPS 2025 · similarity 0.86)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.85)
- Context-Enhanced Memory-Refined Transformer for Online Action Detection (CVPR 2025 · similarity 0.85)
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.
