Human Joint Kinematics Diffusion-Refinement for Stochastic Motion Prediction

Authors: Dong Wei, Huaijiang Sun, Bin Li, Jianfeng Lu, Weiqing Li, Xiaoning Sun, Shengxiang Hu

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on two datasets demonstrate that our model yields the competitive performance in terms of both diversity and accuracy. Extensive experiments show that our model achieves state-of-the-art performance on both Human3.6M and Human Eva-I datasets.
Researcher Affiliation Collaboration 1School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, China 2Tianjin Ai Forward Science and Technology Co., Ltd., Tianjin, China
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes More visualization results can be found in https://github.com/csdwei/Motion Diff.
Open Datasets Yes we evaluate our model on two public benchmark datasets including Human3.6M (Ionescu et al. 2013) and Human Eva-I (Sigal, Balan, and Black 2010).
Dataset Splits No We use 5 subjects (S1, S5, S6, S7, S8) to train the model, and the rest (S9, S11) for evaluation.
Hardware Specification Yes All the experiments are implemented on an NVIDIA RTX 3080 GPU.
Software Dependencies No Our code is in Pytorch (Paszke et al. 2017) and we use ADAM (Kingma and Ba 2015) optimizer.
Experiment Setup Yes For the diffusion network, we use joint embedding layer to upsample the 3D coordinate of human joints from 3 to 32, and then feed it into transformer where the hidden dimension is set to 512. For the refinement network, we use a 12-layers graph convolution network and set the hidden size to 256 in each layer. We set the variance schedules to be β1 = 0.0001 and βK = 0.05, where βk are linearly interpolated (1 < k < K). We train our diffusion model for 1,000 epochs with a batch size of 64 for both datasets.