GCNext: Towards the Unity of Graph Convolutions for Human Motion Prediction
Authors: Xinshun Wang, Qiongjie Cui, Chen Chen, Mengyuan Liu
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on Human3.6M, AMASS, and 3DPW datasets show that, by incorporating unique module-to-network designs, GCNext yields up to 9 lower computational cost than existing GCN methods, on top of achieving state-of-the-art performance. |
| Researcher Affiliation | Collaboration | Xinshun Wang1,2, Qiongjie Cui3, Chen Chen4, Mengyuan Liu2* 1School of Intelligent Systems Engineering, Sun Yat-sen University 2National Key Laboratory of General Artificial Intelligence, Peking University, Shenzhen Graduate School 3Xiaohongshu Inc. 4Center for Research in Computer Vision, University of Central Florida |
| Pseudocode | No | No structured pseudocode or algorithm blocks were found in the paper. |
| Open Source Code | Yes | Our code is available at https://github.com/Bradley Wang0416/GCNext. |
| Open Datasets | Yes | Human3.6M (H3.6M) (Ionescu et al. 2013) comprises 15 types of actions by 7 actors. ... AMASS (Mahmood et al. 2019) combines multiple Mocap datasets unified by SMPL parameterization. ... 3D Pose in the Wild (3DPW) (Von Marcard et al. 2018) includes activities captured from indoor and outdoor scenes. |
| Dataset Splits | Yes | For H3.6M dataset, ... For the validation and test sets, we respectively use subject 11 and subject 5. The remaining 5 subjects for training. ... For AMASS dataset, ... we use AMASS-BMLrub as the test set and split the rest into training and validation sets. |
| Hardware Specification | Yes | The model was trained on RTX 3080 Ti GPU, and the training consumed about 3GB memory and took about 3 hours with batch size of 256. |
| Software Dependencies | No | The paper does not explicitly state specific version numbers for any software dependencies or libraries. |
| Experiment Setup | Yes | The model was trained on RTX 3080 Ti GPU, and the training consumed about 3GB memory and took about 3 hours with batch size of 256. For H3.6M dataset, the model is trained for 85k iterations. The learning rate starts with 0.0006, and drops to 0.000005 after 75k iterations. For AMASS dataset, the model is trained for 115k iterations. The learning rate starts with 0.0003, and drops to 0.000001 after 100k iterations. |