XAGen: 3D Expressive Human Avatars Generation

Authors: Zhongcong XU, Jianfeng Zhang, Jun Hao Liew, Jiashi Feng, Mike Zheng Shou

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that XAGen surpasses state-of-the-art methods in terms of realism, diversity, and expressive control abilities. We conduct extensive experiments on a variety of benchmarks [18, 68, 14, 36], demonstrating the superiority of XAGen over state-of-the-arts in terms of appearance, geometry, and controllability.
Researcher Affiliation Collaboration Zhongcong Xu Show Lab National University of Singapore zhongcongxu@u.nus.edu Jianfeng Zhang Byte Dance jianfengzhang@bytedance.com Jun Hao Liew Byte Dance junhao.liew@bytedance.com Jiashi Feng Byte Dance jshfeng@bytedance.com Mike Zheng Shou Show Lab National University of Singapore mike.zheng.shou@gmail.com
Pseudocode No The paper describes the method using diagrams and mathematical formulas but does not provide pseudocode or algorithm blocks.
Open Source Code No Code and data will be made available at https://showlab.github.io/xagen.
Open Datasets Yes We evaluate the performance of XAGen on four datasets, i.e., Deep Fashion [36], MPV [68], UBC [14], and SHHQ [18].
Dataset Splits No The paper states it uses training data but does not explicitly provide specific proportions or counts for training, validation, and test splits needed for reproduction.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory used for running experiments.
Software Dependencies No The paper mentions using a 'pretrained model [17]' for SMPL-X parameter estimation but does not provide specific version numbers for software dependencies or libraries.
Experiment Setup No The paper describes the components of the loss function and relative sizes of Tri-planes (e.g., Wf = Wh = Wb/2) but does not provide specific hyperparameter values like learning rates, batch sizes, number of epochs, or the numerical values for the weighting factors (λ) in the loss terms.