Research map / Video / Action recognition
Lattice Long Short-Term Memory for Human Action Recognition
Lin Sun, Kui Jia, Kevin Chen, Dit-Yan Yeung, Bertram E. Shi, Silvio Savarese
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 long-short term sampling (ablation) | ConvLSTM spatial network · UCF-101 split 1 | accuracy (%): 77.6 -> 80.4 | Improved |
| add long-short term sampling (ablation) | ConvLSTM temporal network · UCF-101 split 1 | accuracy (%): 79.1 -> 82.3 | Improved |
| add local superposition on memory cell (ablation) | ConvLSTM with sampling, spatial network · UCF-101 split 1 | accuracy (%): 80.4 -> 82.2 | Improved |
| add local superposition on memory cell (ablation) | ConvLSTM with sampling, temporal network · UCF-101 split 1 | accuracy (%): 82.3 -> 86.9 | Improved |
| add shared controlling gates across RGB and flow (ablation) | L2STM spatial network · UCF-101 split 1 | accuracy (%): 82.2 -> 83.2 | Improved |
| add shared controlling gates across RGB and flow (ablation) | L2STM temporal network · UCF-101 split 1 | accuracy (%): 86.9 -> 87.4 | Improved |
| replace ConvLSTM with L2STM | two-stream model · UCF-101 split 1 Human-Object Interaction | accuracy (%): 76.0 -> 86.7 | Improved |
| replace ConvLSTM with L2STM | two-stream model · UCF-101 split 1 Human-Human Interaction | accuracy (%): 89.1 -> 95.4 | Improved |
| replace ConvLSTM with L2STM | two-stream model · UCF-101 split 1 Body Motion Only | accuracy (%): 88.1 -> 88.6 | No clear effect |
| replace ConvLSTM with L2STM | two-stream model · UCF-101 split 1 Pizza Tossing | accuracy (%): 21.2 -> 72.7 | Improved |
| use long-short term sampling instead of fixed stride random clips (ablation) | all LSTM models · UCF-101 | accuracy (%): about 2% gain | Improved |
| replace VideoLSTM with L2STM | LSTM-like action recognizers · UCF-101 mean | accuracy (%): 89.2 -> 93.6 | Improved |
Closest papers by meaning
- Memory-Augmented Temporal Dynamic Learning for Action Recognition (AAAI 2019 · similarity 0.92)
- Regularizing Long Short Term Memory With 3D Human-Skeleton Sequences for Action Recognition (CVPR 2016 · similarity 0.90)
- Unsupervised Learning of Long-Term Motion Dynamics for Videos (CVPR 2017 · similarity 0.90)
- Long-Term Temporal Convolutions for Action Recognition (TPAMI 2017 · similarity 0.90)
- Learning Motion in Feature Space: Locally-Consistent Deformable Convolution Networks for Fine-Grained Action Detection (ICCV 2019 · similarity 0.89)
- A Multi-Stream Bi-Directional Recurrent Neural Network for Fine-Grained Action Detection (CVPR 2016 · similarity 0.89)
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.
