A Large-Scale 3D Face Mesh Video Dataset via Neural Re-parameterized Optimization
Authors: Kim Youwang, Lee Hyun, Kim Sung-Bin, Suekyeong Nam, Janghoon Ju, Tae-Hyun Oh
TMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We assess the fidelity of our dataset by investigating the cross-view vertex distance and the 3D motion stability index. We demonstrate that our dataset contains more spatio-temporally consistent and accurate 3D meshes than the competing datasets built with strong baseline methods. To demonstrate the potential of our dataset, we present two applications: (1) improving the accuracy of a face reconstruction model and (2) learning a generative 3D facial motion prior. These applications highlight that Neu Face-dataset can be further used in diverse applications demanding high-quality and large-scale 3D face meshes. |
| Researcher Affiliation | Collaboration | Kim Youwang EMAIL Department of Electrical Engineering, POSTECH Lee Hyun EMAIL Department of Electrical Engineering, POSTECH Kim Sung-Bin EMAIL Department of Electrical Engineering, POSTECH Suekyeong Nam EMAIL KRAFTON Janghoon Ju EMAIL KRAFTON Tae-Hyun Oh EMAIL Department of Electrical Engineering & Grad. School of AI, POSTECH I-CREATE, Yonsei University |
| Pseudocode | No | The paper describes the Neu Face optimization process in detail across Section 3 and 3.2, and outlines the 'Overall process' in paragraph form, but it does not present any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it use structured, code-like formatting for its procedures. |
| Open Source Code | Yes | Project page: https://kim-youwang.github.io/neuface ... Reproducibility Statement We will make our code and data accessible to the public. |
| Open Datasets | Yes | The Neu Face-dataset provides accurate and spatio-temporally consistent face meshes of existing large-scale 2D face video datasets; MEAD (Wang et al., 2020), Vox Celeb2 (Chung et al., 2018), and Celeb V-HQ (Zhu et al., 2022). |
| Dataset Splits | No | Table 2 shows the MSIL 3D and MSIV 3D averaged over the validation sets. ... We consider the Neu Face holdout test split as the real motion distribution and compute the FD for the generated motions. ... The paper mentions 'validation sets' and a 'holdout test split' but does not provide specific details on the percentages, sample counts, or the methodology used to create these splits for reproducibility. |
| Hardware Specification | No | The paper does not explicitly describe the hardware specifications, such as exact GPU or CPU models, used for conducting the experiments. |
| Software Dependencies | No | The paper mentions several software components and models used, such as FLAME (Li et al., 2017), DECA (Feng et al., 2021), EMOCA (Danecek et al., 2022), FAN (Bulat & Tzimiropoulos, 2017), and Hu Mo R (Rempe et al., 2021), but it does not specify their version numbers or any other general software dependencies with version information required for reproducibility. |
| Experiment Setup | Yes | During fine-tuning, we use an adjusted learning rate, 1 × 10−5, which is ten times smaller than training DECA from scratch. |