Research map / Video / Video segmentation and tracking
Repetition Estimation
Tom F. H. Runia, Cees G. M. Snoek, Arnold W. M. Smeulders
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 wavelet-power motion segmentation instead of full-frame pooling (ablation) | wavelet repetition counter · QUVA Repetition | MAE: 60.8 -> 26.1 | Improved |
| use wavelet-derived segmentation instead of classical motion segmentation (ablation) | wavelet repetition counter · YTSegments | MAE: 13.1 -> 9.4 | Improved |
| use wavelet-derived segmentation instead of two-stream LSTM segmentation (ablation) | wavelet repetition counter · QUVA Repetition | MAE: 38.9 -> 26.1 | Improved |
| combine all six motion maps (oracle selection) instead of one map (ablation) | motion maps · QUVA Repetition | MAE: 42.0 -> 24.1 | Improved |
| use best single map vertical flow vs divergence (ablation) | motion maps · QUVA Repetition | MAE: 77.8 -> 42.0 | Improved |
| replace FlowNet 2.0 with TV-L1 flow (ablation) | wavelet repetition counter · YTSegments | MAE: 9.4 -> 9.8 | No clear effect |
| replace FlowNet 2.0 with EpicFlow (ablation) | wavelet repetition counter · QUVA Repetition | MAE: 26.1 -> 30.8 | Hurt |
| replace Fourier periodogram with wavelet scalogram (ablation) | idealized signals · QUVA Repetition | counting error: 83 out of 100 | Improved |
| add 2x midpoint acceleration to test videos | wavelet repetition counter · YTSegments | MAE: increase of 4.4 | Hurt |
| add 2x midpoint acceleration to test videos | deep learning counting baseline · YTSegments | MAE: rises with 10.8 points | Hurt |
| sum wavelet power of vertical flow and divergence maps (ablation) | bouncing ball with viewpoint change · synthetic Blender video | spectral power continuity | Improved |
| vary spatial sigma from 4 to 2, 8, 16 (ablation) | motion maps · QUVA Repetition and YTSegments | count performance | No clear effect |
Closest papers by meaning
- Real-World Repetition Estimation by Div, Grad and Curl (CVPR 2018 · similarity 0.94)
- Context-Aware and Scale-Insensitive Temporal Repetition Counting (CVPR 2020 · similarity 0.87)
- Repetitive Activity Counting by Sight and Sound (CVPR 2021 · similarity 0.87)
- Counting Out Time: Class Agnostic Video Repetition Counting in the Wild (CVPR 2020 · similarity 0.87)
- Seeing Fast and Slow: Learning the Flow of Time in Videos (ECCV 2026 · similarity 0.86)
- TransRAC: Encoding Multi-Scale Temporal Correlation With Transformers for Repetitive Action Counting (CVPR 2022 · 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.
