Research map / Video / Action recognition
Leveraging temporal contextualization for video action recognition
Minji Kim, Dongyoon Han, Taekyung Kim, Bohyung Han
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 Contextualization (TC) to ViFi-CLIP baseline (ablation) | ViFi-CLIP baseline · Zero-shot (no WE) | top-1 accuracy (All): 67.3 -> 68.0 | Improved |
| add Video-conditional Prompting (VP) (ablation) | ViFi-CLIP baseline · Zero-shot (no WE) | top-1 accuracy (All): 67.3 -> 68.4 | Improved |
| add TC and VP with weight-space ensembling (ablation) | ViFi-CLIP baseline · Zero-shot (WE) | top-1 accuracy (All): 69.0 -> 71.0 | Improved |
| use all patch tokens from all frames as reference (joint space-time) (ablation) | frame-wise baseline · HMDB-51 16-shot | accuracy: 67.1 -> 63.3 | Hurt |
| use context tokens instead of CLS or adjacent-frame tokens as reference (ablation) | frame-wise baseline · SSv2 16-shot | accuracy: 12.0 -> 14.3 | Improved |
| use seed tokens without merging into context tokens (ablation) | TC · Few-shot average | top-1 accuracy (All): 53.5 -> 50.2 | Hurt |
| merge tokens randomly rather than by relevance (ablation) | TC · Few-shot average | top-1 accuracy (All): 53.5 -> 51.2 | Hurt |
| select seed tokens by averaged CLS attention instead of no selection (ablation) | TC · Few-shot average | top-1 accuracy (All): 54.1 -> 54.5 | Improved |
| vary seed token ratio alpha from 0.2 to 0.6 (ablation) | TC · Few-shot average | top-1 accuracy (All): 54.2 to 54.5 | No clear effect |
| add learnable bias between local and global keys with spatial embedding (ablation) | TC · Few-shot average | top-1 accuracy (All): 54.2 -> 54.5 | Improved |
| add learnable prompt vectors without video conditioning (ablation) | ViFi-CLIP baseline · Zero-shot (no WE) | top-1 accuracy (All): 67.3 -> 67.1 | Hurt |
| use VP with context tokens instead of vision-text late-fusion (ablation) | VP · Zero-shot K-600 | top-1 accuracy: 70.9 -> 72.7 | Improved |
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.
- FROSTER — FROSTER: Frozen CLIP is A Strong Teacher for Open-Vocabulary Action Recognition (ICLR 2024) · 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
- Kronecker Mask and Interpretive Prompts are Language-Action Video Learners (ICLR 2025 · similarity 0.89)
- How Much Temporal Long-Term Context is Needed for Action Segmentation? (ICCV 2023 · similarity 0.89)
- Temporal Context Aggregation Network for Temporal Action Proposal Refinement (CVPR 2021 · similarity 0.88)
- OnlineTAS: An Online Baseline for Temporal Action Segmentation (NeurIPS 2024 · similarity 0.87)
- BDC-CLIP: Brownian Distance Covariance for Adapting CLIP to Action Recognition (ICML 2025 · similarity 0.87)
- STOP: Integrated Spatial-Temporal Dynamic Prompting for Video Understanding (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.
