ContactGen: Contact-Guided Interactive 3D Human Generation for Partners

Authors: Dongjun Gu, Jaehyeok Shim, Jaehoon Jang, Changwoo Kang, Kyungdon Joo

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate Contact Gen on the CHI3D dataset, where our method generates physically plausible and diverse poses compared to comparison methods.
Researcher Affiliation Academia Dongjun Gu, Jaehyeok Shim, Jaehoon Jang, Changwoo Kang, Kyungdon Joo* Artifcial Intelligence Graduate School, UNIST {djku1020, jh.shim, erick1997, kangchangwoo, kyungdon}@unist.ac.kr
Pseudocode Yes Algorithm 1: Training; Algorithm 2: Guided Sampling
Open Source Code Yes Source code is available at https://dongjunku.github.io/contactgen.
Open Datasets Yes We basically use CHI3D (Fieraru et al. 2020), which is a 3D motion capture dataset of 8 close human interaction scenarios
Dataset Splits No The paper mentions using the CHI3D dataset and performing pre-processing for training but does not provide specific details on train/validation/test splits (e.g., percentages or sample counts).
Hardware Specification No The paper mentions support from an 'HPC Support Project' but does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for experiments.
Software Dependencies No The paper mentions using SMPL-X representation, Adam optimizer, DDM, and classifier-free guidance, but it does not specify version numbers for any software libraries or frameworks used.
Experiment Setup Yes For the training procedure, noise is sampled from the linear noise scheduler initialized with β0 = 5e 6, βT = 5e 3, and diffusion step T = 1000.