Research map / Video / Video segmentation and tracking
Convolutional Sequence to Sequence Model for Human Dynamics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the convolutional sequence-to-sequence model instead of RRNN | Human motion prediction · Human3.6M average | Euler angle error: 0.43 -> 0.38; 0.75 -> 0.68; 1.12 -> 1.01; 1.27 -> 1.13; 1.90 -> 1.77 | Helped |
| Use the convolutional sequence-to-sequence model instead of RRNN | Human motion prediction · CMU Motion Capture average | Euler angle error: 0.38 -> 0.31; 0.62 -> 0.52; 1.02 -> 0.86; 1.17 -> 0.99; 1.67 -> 1.55 | Helped |
| Add a long-term encoder (ablation) | Convolutional sequence-to-sequence model · Human3.6M average | Euler angle error: 0.41 -> 0.38; 0.72 -> 0.68; 1.05 -> 1.01; 1.17 -> 1.13; 1.88 -> 1.77 | Helped |
| Use a 2 × 7 spatial rectangular kernel instead of a 4 × 4 square kernel (ablation) | Convolutional encoding module · Human3.6M | Euler angle error: 0.41, 0.72, 1.05, 1.16, 1.80 -> 0.38, 0.68, 1.01, 1.13, 1.77 | Helped |
| Use a 2 × 7 spatial rectangular kernel instead of a 7 × 2 temporal rectangular kernel (ablation) | Convolutional encoding module · Human3.6M | Euler angle error: 0.40, 0.71, 1.05, 1.17, 1.79 -> 0.38, 0.68, 1.01, 1.13, 1.77 | Helped |
| Add adversarial regularization during training (ablation) | Convolutional sequence-to-sequence model · Human3.6M | testing error | Helped |
| Increase short-term window size from 5 to 10 (ablation) | Convolutional sequence-to-sequence model · Human3.6M, long-term prediction | testing error | Helped |
| Increase short-term window size from 10 to 20 (ablation) | Convolutional sequence-to-sequence model · Human3.6M | testing error | No clear effect |
| Replace the convolutional long-term encoder with an LSTM encoder (ablation) | Convolutional sequence-to-sequence model · Human3.6M | prediction error over time | Hurt |
Closest papers by meaning
- Deep Representation Learning for Human Motion Prediction and Classification (CVPR 2017 · similarity 0.90)
- On Human Motion Prediction Using Recurrent Neural Networks (CVPR 2017 · similarity 0.89)
- Decomposing Motion and Content for Natural Video Sequence Prediction (ICLR 2017 · similarity 0.87)
- A Neural Temporal Model for Human Motion Prediction (CVPR 2019 · similarity 0.87)
- Understanding image motion with group representations (ICLR 2018 · similarity 0.86)
- Trajectory Convolution for Action Recognition (NeurIPS 2018 · similarity 0.86)
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.