Research map / Video / Action recognition
K-Centered Patch Sampling for Efficient Video Recognition
Seong Hyeon Park, Jihoon Tack, Byeongho Heo, Jung-Woo Ha, Jinwoo Shin
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 |
|---|---|---|---|
| sample random patches instead of random frames | DeiT-base video extension · Something-Something v2 | Top-1 accuracy: 58.95 -> 61.64 | Improved |
| use K-centered sampling instead of random patch sampling | DeiT-base video extension · Something-Something v2 | Top-1 accuracy: 61.64 -> 64.31 | Improved |
| use K-centered patch sampling instead of frame-based | Motionformer · Kinetics-400 | Top-1 accuracy: 71.42 -> 74.85 | Improved |
| use K-centered patch sampling instead of frame-based | ViT · Something-Something v2 | Top-1 accuracy: 62.50 -> 63.83 | Improved |
| use K-centered patch sampling instead of frame-based | XViT · Something-Something v2 | Top-1 accuracy: 62.40 -> 62.78 | Improved |
| use K-centered patch sampling instead of frame-based | TimeSformer · Kinetics-400 | Top-1 accuracy: 77.95 -> 77.98 | No clear effect |
| use K-centered patch sampling instead of frame-based | TimeSformer · Something-Something v2 | Top-1 accuracy: 63.76 -> 63.81 | No clear effect |
| choose patches using RGB instead of grayscale distance (ablation) | ViT with K-centered sampling · Kinetics-200 | Top-1 accuracy: 81.94 -> 82.90 | Improved |
| add 3 complete frames via hybrid sampling (ablation) | ViT with K-centered sampling · Kinetics-200 | Top-1 accuracy: 81.56 -> 82.90 | Improved |
| add hybrid frames to patch sampling (ablation) | Motionformer with K-centered sampling · Kinetics-200 | Top-1 accuracy | No clear effect |
| raise distance coefficients from ws=1, wt=1 to ws=10, wt=10 (ablation) | ViT with K-centered sampling · Kinetics-200 | Top-1 accuracy: 82.90 -> 81.56 | Hurt |
| train with K-centered sampling instead of frame-based, test with 50% patches dropped (ablation) | ViT · Kinetics-200 | Top-1 accuracy: 75.64 -> 79.28 | Improved |
Closest papers by meaning
- MGSampler: An Explainable Sampling Strategy for Video Action Recognition (ICCV 2021 · similarity 0.85)
- ViViT: A Video Vision Transformer (ICCV 2021 · similarity 0.85)
- AsymRnR: Video Diffusion Transformers Acceleration with Asymmetric Reduction and Restoration (ICML 2025 · similarity 0.83)
- Hierarchical Patch Diffusion Models for High-Resolution Video Generation (CVPR 2024 · similarity 0.83)
- PatchVSR: Breaking Video Diffusion Resolution Limits with Patch-wise Video Super-Resolution (CVPR 2025 · similarity 0.83)
- vid-TLDR: Training Free Token Merging for Light-weight Video Transformer (CVPR 2024 · similarity 0.82)
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.
