HVIS: A Human-like Vision and Inference System for Human Motion Prediction
Authors: Kedi Lyu, Haipeng Chen, Zhenguang Liu, Yifang Yin, Yukang Lin, Yingying Jiao
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that our method achieves new state-of-the-art performance, significantly outperforming existing methods by 19.8% on Human3.6M, 15.7% on CMU Mocap, and 11.1% on G3D. The paper includes dedicated sections for 'Experiments', 'Datasets and Experimental Settings', 'Comparison with Existing Methods', and 'Ablation Experiments', presenting quantitative results in tables and visual comparisons. |
| Researcher Affiliation | Academia | 1College of Computer Science and Technology, Jilin University, Changchun, China 2Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University.Changchun, China 3The State Key Laboratory of Blockchain and Data Security, Zhejiang University, Hangzhou, China 4Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security, Hangzhou, China 5Institute for Infocomm Research (I2R), A*STAR, Singapore 6Tsinghua Shenzhen International Graduate School, Nanshan District, Shenzhen, China EMAIL, EMAIL, liuzhenguang2008 @gmail.com, yin EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes its methodology using descriptive text and mathematical formulations (e.g., equations 1-12) but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Specific experimental proofs are available in our open-source library. |
| Open Datasets | Yes | To testify the effectiveness and robustness of our proposed model, three large benchmark datasets Human 3.6 Million (H3.6M), CMU Mo Cap (CMU), and G3D are engaged. |
| Dataset Splits | No | For the CMU dataset, the paper states: 'In general, these samples are split into a training set and a test set in the experiment.' However, it does not provide specific percentages, sample counts, or a detailed splitting methodology for any of the datasets (H3.6M, CMU, G3D) used in the experiments. While these are common benchmark datasets, the paper does not explicitly state the splits used for reproduction within its text. |
| Hardware Specification | Yes | We build our model on the Py Torch with a NVIDIA 3090Ti GPU. |
| Software Dependencies | No | The paper mentions 'Py Torch' as the framework used but does not specify its version or any other software dependencies with version numbers. |
| Experiment Setup | Yes | We build our model on the Py Torch with a NVIDIA 3090Ti GPU. The Adam Optimizer is utilized with a learning rate of 0.001. The TIU has three blocks. In each block, kernel-size is 3, dropout rate is 0.1 and dilation rate is 2i 1 in each layer i (i < 4).The linear has 256 units. For LTF, the hidden unit size is 256. The critic is a three-layer FCN with 256 units. In the DLN, a three-block TCN is used with 4 layers and 0.2 dropout rate in each block. The lengths of the observed sequence and the predicted sequence are set to 25 frames. Note that different datasets and different actions are trained independently in our method. |