Learning to Decouple the Lights for 3D Face Texture Modeling

Authors: Tianxin Huang, Zhenyu Zhang, Ying Tai, Gim Hee Lee

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

Reproducibility Variable Result LLM Response
Research Type Experimental According to experiments on both single images and video sequences, we demonstrate the effectiveness of our approach in modeling facial textures under challenging illumination affected by occlusions. Our codes would be open sourced at https://github.com/Tianxinhuang/De Face.git.
Researcher Affiliation Academia Tianxin Huang1 Zhenyu Zhang2 Ying Tai2 Gim Hee Lee1 1School of Computing, National University of Singapore 2Nanjing University
Pseudocode Yes Algorithm 1 Training Process
Open Source Code Yes Our codes would be open sourced at https://github.com/Tianxinhuang/De Face.git.
Open Datasets Yes For our experiments, we utilize two datasets: Voxceleb2 [8] and Celeb AMask-HQ [29, 24]. Vox Celeb2 [8] is a diverse dataset encompassing numerous videos collected from interviews, movies and videos, where the same person may have multiple separate videos. Celeb AMask-HQ [29, 24] is a large scale face image dataset with fine attributes annotation and high resolution, widely used in face editing and generation.
Dataset Splits No The paper mentions training and testing but does not explicitly specify a separate validation dataset split with percentages or counts.
Hardware Specification Yes We conduct all experiments on a Nvidia 2080ti GPU with a 2.9Ghz i5-9400 CPU.
Software Dependencies No The paper mentions 'Py Torch version' and 'Adam optimizer [19]' but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes Detailed hyper-parameter settings can be found in our supplementary. Table 6: Hyper-parameter settings. n is the number of initial light conditions presented in Fig. 2. w1 w7 2e3, 1e-3, 1.5e2, 0.5, 25, 2e3, 2.0, 1.0 Landmark Mediapipe iter0 iter3 100, 2000, 400, 200 ϵ 0.17 n 5