Research map / Video / Action recognition
Chained Multi-Stream Networks Exploiting Pose, Motion, and Appearance for Action Classification and Detection
Mohammadreza Zolfaghari, Gabriel L. Oliveira, Nima Sedaghat, Thomas Brox
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 |
|---|---|---|---|
| fuse RGB and optical flow with the Markov chain instead of feature concatenation (ablation) | RGB+OF two-stream C3D · J-HMDB (split 1) | accuracy: 62.7% -> 72.8% | Improved |
| fuse RGB, flow and pose with the chain instead of concatenation (ablation) | three-stream C3D · HMDB51 (split 1) | accuracy: 57.5% -> 62.1% | Improved |
| fuse RGB, flow and pose with the chain instead of concatenation (ablation) | three-stream C3D · J-HMDB (split 1) | accuracy: 70.2% -> 79.1% | Improved |
| fuse RGB, flow and pose with the chain instead of concatenation (ablation) | three-stream C3D · UCF101 (split 1) | accuracy: 89.1% -> 90.4% | Improved |
| add the pose stream to the chained RGB+OF network (ablation) | chained C3D streams · J-HMDB (split 1) | accuracy: 72.8% -> 79.1% | Improved |
| add the pose stream to the chained RGB+OF network (ablation) | chained C3D streams · UCF101 (split 1) | accuracy: 88.9% -> 90.4% | Improved |
| add multi-granular temporal fusion (MG) (ablation) | three-stream chained network · HMDB51 (split 1) | accuracy: 62.1% -> 71.1% | Improved |
| use ground-truth flow and pose instead of estimated ones (ablation) | three-stream chained network · J-HMDB (split 1) | accuracy: 79.1% -> 83.2% | Improved |
| order the chain pose, then flow, then RGB rather than RGB first (ROP) (ablation) | chained network · HMDB51 (split 1) | clip accuracy: 54.1% -> 60.0% | Improved |
| fuse at FC6 instead of FC7 (ablation) | chained network · J-HMDB (split 1) | accuracy: 73.9% -> 79.1% | Improved |
| fuse at FC6 instead of FC7 (ablation) | chained network · UCF101 (split 1) | accuracy: 89.8% -> 89.6% | No clear effect |
| increase clip length from 4 to 16 frames (ablation) | RGB stream · J-HMDB (RGB) | accuracy: 44.8% -> 60.8% | Improved |
Closest papers by meaning
- Action Recognition With Spatial-Temporal Discriminative Filter Banks (ICCV 2019 · similarity 0.89)
- MARS: Motion-Augmented RGB Stream for Action Recognition (CVPR 2019 · similarity 0.89)
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.88)
- Convolutional Two-Stream Network Fusion for Video Action Recognition (CVPR 2016 · similarity 0.88)
- Deep Insights into Convolutional Networks for Video Recognition (IJCV 2019 · similarity 0.88)
- ActionVLAD: Learning Spatio-Temporal Aggregation for Action Classification (CVPR 2017 · 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.
