Research map / Video / Action recognition
Counting Out Time: Class Agnostic Video Repetition Counting in the Wild
Debidatta Dwibedi, Yusuf Aytar, Jonathan Tompson, Pierre Sermanet, Andrew Zisserman
Check 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 RepNet rather than P-MUCOS for periodicity detection | periodicity detection · PERTUBE | F1: 0.77 -> 0.820 | Improved |
| Add the temporal self-similarity matrix bottleneck (ablation) | RepNet trained on synthetic repetitions · QUVA | MAE: 1.2853 -> 0.1035 | Improved |
| Add the temporal self-similarity matrix bottleneck (ablation) | RepNet trained on synthetic repetitions · Countix val | MAE: 1.1671 -> 0.3100 | Improved |
| Train on synthetic repetitions rather than Countix labels (ablation) | RepNet with TSM · QUVA | MAE: 0.3225 -> 0.1035 | Improved |
| Add Countix real training videos to synthetic data (ablation) | RepNet with TSM · QUVA | OBO: 0.17 -> 0.15 | Improved |
| Replace the transformer period predictor with LSTM (ablation) | RepNet · Countix val | MAE: 0.3100 -> 0.6895 | Hurt |
| Increase camera motion augmentation from none to fraction 0.75 (ablation) | synthetic-data RepNet · QUVA | MAE: 0.7178 -> 0.1035 | Improved |
| Exclude Countix classes from the synthetic training pool (ablation) | RepNet · QUVA | MAE: 0.1035 -> 0.1181 | No clear effect |
| Remove ImageNet pretraining (ablation) | RepNet · QUVA | MAE: 0.1035 -> 0.3097 | Hurt |
| Remove scale augmentation (ablation) | synthetic-data RepNet · Countix val | MAE: 0.3100 -> 0.5751 | Hurt |
| Use 32 rather than 64 input frames (ablation) | RepNet · QUVA | MAE: 0.1035 -> 0.1407 | Hurt |
| Remove 3D convolution (ablation) | RepNet · QUVA | MAE: 0.1035 -> 0.1198 | Hurt |
Closest papers by meaning
- Context-Aware and Scale-Insensitive Temporal Repetition Counting (CVPR 2020 · similarity 0.93)
- Efficient Action Counting with Dynamic Queries (IJCV 2026 · similarity 0.89)
- TransRAC: Encoding Multi-Scale Temporal Correlation With Transformers for Repetitive Action Counting (CVPR 2022 · similarity 0.89)
- CountLLM: Towards Generalizable Repetitive Action Counting via Large Language Model (CVPR 2025 · similarity 0.89)
- FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.83)
- ChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation (NeurIPS 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.
