Research map / Video / Action recognition
Regularization on Spatio-Temporally Smoothed Feature for Action Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 RMS before all BNs | SlowOnly-34 · Mini-Kinetics | Top1 Acc.(%): 74.7±0.33 -> 77.0±0.31 | Helped |
| Use Gaussian filtering for RMS (ablation) | SlowOnly-34 · Mini-Kinetics | Top1 Acc.(%): 74.7±0.33 -> 77.5±0.16 | Helped |
| Add RMS before the last BN (ablation) | SlowOnly-50 · Mini-Kinetics | Top1 Acc.(%): 77.5±0.85 -> 78.6±0.33 | Helped |
| Apply RMS to mean component instead of full feature or residual (ablation) | SlowOnly-34 · Mini-Kinetics | Top1 Acc.(%): RS on x 75.4±0.40; RS on r 75.6±0.77; RS on x̄ (RMS) 77.0±0.31 | Helped |
| Add RMS | SlowOnly · Kinetics-400 | Top1 Acc.(%): 73.0 -> 74.2 | Helped |
| Add RMS | ip-CSN · Kinetics-400 | Top1 Acc.(%): 69.7 -> 70.5 | Helped |
| Add RMS | SlowFast · Kinetics-400 | Top1 Acc.(%): 75.0 -> 76.3 | Helped |
| Add RMS | SlowFast-50 · Something-Something-v2 | Top1 Acc.(%): 59.5 -> 61.2 | Helped |
| Add RMS | ResNet-110 · CIFAR-100 | accuracy: 74.49∗ -> 77.17∗ | Helped |
| Add RMS | ResNet-50 · ImageNet | Top1 Acc.(%): 76.81 -> 77.62 | Helped |
| Use U(−0.5, 2.5) rather than the default normal sampling distribution for α (ablation) | SlowOnly-34 + RMS · Mini-Kinetics | Top1 Acc.(%): 77.0±0.31 -> 74.5±0.58 | Hurt |
| Use Gaussian rather than mean filtering in RMS (ablation) | SlowOnly-50 · Mini-Kinetics | Top1 Acc.(%): 78.6±0.33 -> 78.3±0.33 | No clear effect |
Closest papers by meaning
- Dynamic Normalization and Relay for Video Action Recognition (NeurIPS 2021 · similarity 0.86)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.85)
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.85)
- Smooth Regularization for Efficient Video Recognition (NeurIPS 2025 · similarity 0.84)
- More Is Less: Learning Efficient Video Representations by Big-Little Network and Depthwise Temporal Aggregation (NeurIPS 2019 · similarity 0.84)
- Slow and Steady Feature Analysis: Higher Order Temporal Coherence in Video (CVPR 2016 · similarity 0.84)
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.