Research map / Video / Action recognition
Spatiotemporal Feature Residual Propagation for Action Prediction
He Zhao, Richard P. Wildes
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 |
|---|---|---|---|
| propagate residuals instead of raw features (ablation) | RGN without Kalman filter · JHMDB21 split 1 (g=0.2) | accuracy (%): 70.9 -> 77.4 | Improved |
| propagate residuals instead of raw features (ablation) | RGN with Kalman filter · JHMDB21 split 1 (g=0.2) | accuracy (%): 74.4 -> 78.3 | Improved |
| add Kalman filter correction (ablation) | RGN on residuals · JHMDB21 split 1 (g=0.2) | accuracy (%): 77.4 -> 78.3 | Improved |
| add Kalman filter correction (ablation) | RGN on raw features · JHMDB21 split 1 (g=0.2) | accuracy (%): 70.9 -> 74.4 | Improved |
| add Kalman filter correction (ablation) | ConvLSTM on residuals · JHMDB21 split 1 (g=0.2) | accuracy (%): 76.8 -> 77.1 | Improved |
| propagate residuals instead of raw features (ablation) | ConvLSTM · JHMDB21 split 1 (g=0.2) | accuracy (%): 71.1 -> 76.8 | Improved |
| use RGN instead of ConvLSTM for propagation (residuals + KF) (ablation) | RGN-Res-KF · JHMDB21 split 1 (g=0.2) | accuracy (%): 77.1 -> 78.3 | Improved |
| propagate features into the future vs classify observed frames only (ablation) | TSN adapted to JHMDB21 · JHMDB21 (g=0.2) | accuracy (%): 74 -> 78 | Improved |
| use 28x28x192 layer instead of 56x56x64 (ablation) | RGN-KF · BIT (g=0.2) | accuracy (%): 35.94 -> 46.09 | Improved |
| use 28x28x192 layer instead of 7x7x1024 (ablation) | RGN-KF · BIT (g=0.2) | accuracy (%): 42.06 -> 46.09 | Improved |
| use 28x28x192 layer instead of 56x56x64 (ablation) | RGN-KF · JHMDB21 (g=0.2) | accuracy (%): 75.74 -> 78.30 | Improved |
| vary intermediate feature layer on UCF101 (ablation) | RGN-KF · UCF101 (g=0.2) | accuracy (%): 84.94 to 85.16 | No clear effect |
Closest papers by meaning
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.89)
- Learning Visual Feature-Based World Models via Residual Latent Action (NeurIPS 2026 · similarity 0.88)
- STEP: Spatio-Temporal Progressive Learning for Video Action Detection (CVPR 2019 · similarity 0.88)
- A Closer Look at Spatiotemporal Convolutions for Action Recognition (CVPR 2018 · similarity 0.88)
- A Large-scale Study of Spatiotemporal Representation Learning with a New Benchmark on Action Recognition (ICCV 2023 · similarity 0.88)
- What Have We Learned From Deep Representations for Action Recognition? (CVPR 2018 · 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.
