Research map / Video / Action recognition
Learning Motion in Feature Space: Locally-Consistent Deformable Convolution Networks for Fine-Grained Action Detection
Khoi-Nguyen C. Mac, Dhiraj Joshi, Raymond A. Yeh, Jinjun Xiong, Rogerio S. Feris, Minh N. Do
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 the local coherency constraint to deformable convolution (ablation) | LCDC · 50 Salads (split 1, mid-level, ablation) | frame-wise accuracy: 72.25 -> 73.77 | Improved |
| add the local coherency constraint (deformation parameters) (ablation) | LCDC · 50 Salads (split 1, mid-level, ablation) | deformable parameters: 995.5K -> 27.7K | Improved |
| use LCDC features instead of SpatialCNN with ED-TCN | ED-TCN · 50 Salads (mid-level) | F1@10: 68.0 -> 73.75 | Improved |
| use LCDC features instead of SpatialCNN with ED-TCN | ED-TCN · 50 Salads (eval-level) | frame-wise accuracy: 73.4 -> 78.90 | Improved |
| use LCDC features instead of SpatialCNN with ED-TCN | ED-TCN · GTEA | F1@10: 72.2 -> 75.39 | Improved |
| use LCDC features instead of SpatialCNN with ED-TCN | ED-TCN · GTEA | frame-wise accuracy: 64.0 -> 65.34 | Improved |
| use LCDC features alone instead of SpatialCNN features | LCDC · 50 Salads (mid-level) | frame-wise accuracy: 54.9 -> 67.27 | Improved |
| add multiple averaged feature frames to appearance stream (ablation) | NaiveAppear · 50 Salads (split 1, mid-level, ablation) | frame-wise accuracy: 68.45 -> 71.52 | Improved |
| replace two-stream with optical flow by learned deformation (LCDC) (ablation) | TwoStreamNet · 50 Salads (split 1, mid-level, ablation) | frame-wise accuracy: 71.82 -> 73.77 | Improved |
| use LCDC features instead of SpatialCNN features at GTEA, no long-temporal model | LCDC · GTEA | frame-wise accuracy: 54.1 -> 55.32 | Mixed |
| use LCDC features with ST-CNN rather than ED-TCN | LCDC · 50 Salads (mid-level) | F1@10: 73.75 -> 60.01 | Hurt |
Closest papers by meaning
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.93)
- Segmental Spatiotemporal CNNs for Fine-Grained Action Segmentation (ECCV 2016 · similarity 0.91)
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · similarity 0.91)
- Graph Convolutional Module for Temporal Action Localization in Videos (TPAMI 2021 · similarity 0.91)
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.90)
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.90)
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.
