Dual-Diffusion for Binocular 3D Human Pose Estimation

Authors: Xiaoyue Wan, Zhuo Chen, Bingzhi Duan, Xu Zhao

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments validate the effectiveness of our Dual-Diffusion in 2D refinement and 3D estimation. and To validate the efficacy of our Dual-Diffusion model in denoising both 2D and 3D poses, we conducted experiments on the binocular H36M [13] and MHAD [27] datasets, utilizing only 2-view camera pairs.
Researcher Affiliation Academia Xiaoyue Wan Zhuo Chen Bingzhi Duan Xu Zhao Department of Automation Shanghai Jiao Tong University {sherrywaan, chzh9311, Duan Bingzhi, zhaoxu}@sjtu.edu.cn
Pseudocode No The paper uses figures like Figure 2 and Figure 9 to illustrate the model and training process, but these are high-level diagrams and flowcharts, not detailed pseudocode or algorithm blocks. The procedural steps are described in paragraph text within the 'Training Details' and 'Inference' sections.
Open Source Code Yes The code and models are available at https://github.com/sherrywan/Dual-Diffusion.
Open Datasets Yes We conducted experiments on the binocular H36M [13] and MHAD [27] datasets, utilizing only 2-view camera pairs.
Dataset Splits No The paper mentions training on MHAD and H36M and evaluating on testing sets, but it does not specify explicit train/validation/test splits, such as percentages or sample counts for each subset, nor does it refer to predefined standard splits with citations.
Hardware Specification Yes All experiments are conducted on Ge Force RTX 2080 Ti.
Software Dependencies No The Dual-Diffusion is implemented by Py Torch [28] using Adam optimizer with learning rate 0.00002 and other parameters are all default. However, a specific version number for PyTorch is not provided.
Experiment Setup Yes The Dual-Diffusion is implemented by Py Torch [28] using Adam optimizer with learning rate 0.00002 and other parameters are all default. and For the experiments described below, we set T = 25 and K = 1 according to the ablation study in Sec. 4.3 and Appendix D.3. and We train our Dual-Diffusion in MHAD [27] for 80 epochs, and in H36M [13] for 40 epochs, respectively.