Research map / Video / Action recognition
ActionVLAD: Learning Spatio-Temporal Aggregation for Action Classification
Rohit Girdhar, Deva Ramanan, Abhinav Gupta, Josef Sivic, Bryan Russell
Check 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 |
|---|---|---|---|
| Train the ActionVLAD aggregation together with preceding network layers rather than keeping the initialized aggregation fixed (ablation) | ActionVLAD on HMDB51 split 1 · HMDB51 split 1 | accuracy: 44.9 -> 51.2 | Improved |
| Use K=1 cluster instead of K=64 (ablation) | ActionVLAD · HMDB51 split 1 RGB | accuracy: 51.2% -> 43.2% | Hurt |
| Aggregate conv5_3 features instead of conv4_3 or fc7 (ablation) | ActionVLAD · HMDB51 split 1 | accuracy: 45.0 -> 51.2; 43.3 -> 51.2 | Improved |
| Replace ActionVLAD with average pooling over conv5_3 features (ablation) | VGG-16 two-stream architecture · HMDB51 split 1 RGB | accuracy: 51.2 -> 41.6 | Hurt |
| Replace ActionVLAD with max pooling over conv5_3 features (ablation) | VGG-16 two-stream architecture · HMDB51 split 1 RGB | accuracy: 51.2 -> 41.5 | Hurt |
| Fuse appearance and motion by concatenating features before one ActionVLAD layer rather than separately late-fusing their representations (ablation) | ActionVLAD · HMDB51 split 1 | validation accuracy: 66.9 -> 56.0 | Hurt |
| Pool appearance and motion together in one ActionVLAD layer rather than separately late-fusing their representations (ablation) | ActionVLAD · HMDB51 split 1 | validation accuracy: 66.9 -> 64.8 | Hurt |
| Use five crops per frame instead of a single center crop at test time (ablation) | ActionVLAD · HMDB51 | accuracy: 66.3 -> 66.9 | Improved |
| Use ActionVLAD rather than the two-stream baseline | VGG-16 action classification · HMDB51 | accuracy: 58.5 -> 66.3 | Improved |
| Add iDT score fusion to ActionVLAD | ActionVLAD (VGG-16) · Charades | mAP: 17.6 -> 21.0 | Improved |
Closest papers by meaning
- Deep Insights into Convolutional Networks for Video Recognition (IJCV 2019 · similarity 0.89)
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.89)
- Convolutional Two-Stream Network Fusion for Video Action Recognition (CVPR 2016 · similarity 0.89)
- What Have We Learned From Deep Representations for Action Recognition? (CVPR 2018 · similarity 0.89)
- Sympathy for the Details: Dense Trajectories and Hybrid Classification Architectures for Action Recognition (ECCV 2016 · similarity 0.89)
- Chained Multi-Stream Networks Exploiting Pose, Motion, and Appearance for Action Classification and Detection (ICCV 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.
