Research map / Video / Action recognition
StNet: Local and Global Spatial-Temporal Modeling for Action Recognition
Read the paperCheck 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 1D BatchNorm (ablation) | TXB · Kinetics400 feature-sequence experiment | Top-1: 73.55 -> 74.62 | Helped |
| add channel-wise 1D convolution (ablation) | TXB · Kinetics400 feature-sequence experiment | Top-1: 74.14 -> 74.62 | Helped |
| add temporal-wise 1D convolution (ablation) | TXB · Kinetics400 feature-sequence experiment | Top-1: 74.06 -> 74.62 | Helped |
| add the short branch (ablation) | TXB · Kinetics400 feature-sequence experiment | Top-1: 74.33 -> 74.62 | Helped |
| add the long branch (ablation) | TXB · Kinetics400 feature-sequence experiment | Top-1: 74.21 -> 74.62 | Helped |
| replace TXB with ordinary temporal convolution | temporal feature-sequence modeling · Kinetics400 feature-sequence experiment | Top-1: 74.62 -> 74.28 | Hurt |
| replace TXB with LSTM | temporal feature-sequence modeling · Kinetics400 feature-sequence experiment | Top-1: 74.62 -> 73.21 | Hurt |
| replace TXB with GRU | temporal feature-sequence modeling · Kinetics400 feature-sequence experiment | Top-1: 74.62 -> 73.66 | Hurt |
| add super-images formed from N=5 consecutive frames (ablation) | StNet · Kinetics600 validation | Top-1: 72.2 -> 74.2 | Helped |
| add temporal modeling blocks after the super-image model (ablation) | StNet · Kinetics600 validation | Top-1: 74.2 -> 76.0 | Helped |
| add TXB after super-images and temporal modeling blocks (ablation) | StNet · Kinetics600 validation | Top-1: 76.0 -> 76.3 | Helped |
Closest papers by meaning
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.90)
- ACTION-Net: Multipath Excitation for Action Recognition (CVPR 2021 · similarity 0.89)
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · similarity 0.89)
- Collaborative Spatiotemporal Feature Learning for Video Action Recognition (CVPR 2019 · similarity 0.89)
- Are Spatial-Temporal Graph Convolution Networks for Human Action Recognition Over-Parameterized? (CVPR 2025 · similarity 0.88)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.88)
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.