ID-Sculpt: ID-aware 3D Head Generation from Single In-the-wild Portrait Image

Authors: Jinkun Hao, Junshu Tang, Jiangning Zhang, Ran Yi, Yijia Hong, Moran Li, Weijian Cao, Yating Wang, Chengjie Wang, Lizhuang Ma

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In the experimental section, we start with the implementation details, followed by quantitative and qualitative comparisons between our method and others. Finally, we analyze the role of each module in our method.
Researcher Affiliation Collaboration Jinkun Hao1, Junshu Tang1, Jiangning Zhang2, Ran Yi1*, Yijia Hong1, Moran Li2, Weijian Cao2, Yating Wang1, Chengjie Wang1, 2, Lizhuang Ma1* 1Shanghai Jiao Tong University 2Youtu Lab, Tencent
Pseudocode No The paper describes the method using mathematical formulations and textual descriptions but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes Project page https://jinkun-hao.github.io/IDSculpt/
Open Datasets Yes Datasets. We conducted experiments on a subset of facial images from the FFHQ (Karras, Laine, and Aila 2019) dataset, a high-quality in-the-wild facial dataset known for its rich diversity in age, ethnicity, and image backgrounds, as well as significant variations in facial attributes.
Dataset Splits No The paper mentions using a subset of the FFHQ dataset but does not provide specific details on training, testing, or validation splits.
Hardware Specification Yes Our experiments are conducted on a single V100 GPU, with a batch size set to 1.
Software Dependencies Yes Our ID-Sculpt is built upon the open-source project Three Studio (Guo et al. 2023). In the geometry stage, we use SD1.5 (Rombach et al. 2022) as the base diffusion model. During the texture generation stage, we employ Realistic Vision 4.0 as the base diffusion model.
Experiment Setup Yes In the geometry generation stage, we perform 5,000 optimization iterations. During the texture generation stage, we first perform texture inpainting from 15 ordered viewpoints, followed by 400 steps of texture refinement. Our experiments are conducted on a single V100 GPU, with a batch size set to 1.