FlexiTex: Enhancing Texture Generation via Visual Guidance

Authors: Dadong Jiang, Xianghui Yang, Zibo Zhao, Sheng Zhang, Jiaao Yu, Zeqiang Lai, Shaoxiong Yang, Chunchao Guo, Xiaobo Zhou, Zhihui Ke

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct comprehensive studies and analyses involving numerous 3D objects from various sources to demonstrate the effectiveness of Flexi Texin texture generation, as in Fig. 1. We summarize our contributions as follows: [...] Experiments Setup Implementation Details. Our experiments are conducted on an NVIDIA A100 GPU. [...] Quantitative Analysis [...] Qualitative Analysis [...] Ablation Studies
Researcher Affiliation Collaboration Dadong Jiang1,2, Xianghui Yang2, Zibo Zhao2, Sheng Zhang2, Jiaao Yu2, Zeqiang Lai2, Shaoxiong Yang2, Chunchao Guo2*, Xiaobo Zhou1 , Zhihui Ke1 1Tianjin University 2Tencent Hunyuan EMAIL, EMAIL
Pseudocode No The paper describes the Visual Guidance Enhancement module and the Direction-Aware Adaptation module in text, but does not include any structured pseudocode or algorithm blocks.
Open Source Code No Project https://patrickddj.github.io/Flexi Tex/ The provided link is to a project page, not explicitly a source code repository, and there is no unambiguous statement about releasing code.
Open Datasets Yes Dataset. We collect 60 meshes with corresponding text prompts and 60 meshes with corresponding image prompts to evaluate the text-to-texture and image-to-texture generation ability, respectively. These meshed are randomly sampled from Objaverse (Deitke et al. 2022), Objaverse XL (Deitke et al. 2023) and Shape Net (Chang et al. 2015).
Dataset Splits No The paper states that 60 meshes with text prompts and 60 meshes with image prompts are collected from Objaverse, Objaverse XL, and Shape Net, but it does not specify how these meshes are split into training, validation, or test sets for reproducibility.
Hardware Specification Yes Experiments Setup Implementation Details. Our experiments are conducted on an NVIDIA A100 GPU.
Software Dependencies No The paper mentions using DDIM as a sampler, Pytorch3D for rendering, Control Net, Stable Diffusion, and IP-Adaptor, but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes Implementation Details. Our experiments are conducted on an NVIDIA A100 GPU. For denoising epoch, we use DDIM (Song, Meng, and Ermon 2020) as the sampler. We set the number of iterations to 30 steps, the CFG scale (classifier-free guidance scale) for Direction-Aware Adaptation to 12, and the scale of Visual Guidance Enhancement to 0.6. Texture warping for latent views is used in the first 24 steps. We sample 8 views for a mesh, and the elevations and azimuths are ( 180 , 15 ), ( 120 , 15 ), ( 60 , 15 ), (0 , 15 ), (60 , 15 ), (120 , 15 ), ( 180 , 45 ), (0 , 45 ).