IPVTON: Image-based 3D Virtual Try-on with Image Prompt Adapter
Authors: Xiaojing Zhong, Zhonghua Wu, Xiaofeng Yang, Guosheng Lin, Qingyao Wu
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive qualitative and quantitative experiments demonstrate that IPVTON outperforms previous methods in image-based 3D virtual try-on tasks, excelling in both geometry and texture. |
| Researcher Affiliation | Collaboration | 1School of Software Engineering, South China University of Technology, China 2Nanyang Technological University, Singapore 3Sense Time Research, Singapore 4Peng Cheng Laboratory, China |
| Pseudocode | No | The paper describes the method using mathematical formulations and descriptive text, but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statements about the release of source code or links to a code repository. |
| Open Datasets | Yes | We select 12 full-body, frontfacing human images of different individuals from the Deep Fashion dataset (Shen et al. 2021). For each human image, we choose two garment templates from the VITONHD dataset (Choi et al. 2021), covering various types such as tank tops, short sleeves, and long sleeves. |
| Dataset Splits | No | The paper states, 'Our method does not require paired human and clothing images for training.' It mentions selecting human images from Deep Fashion and garment templates from VITONHD for generating results, and '8 sets of human images with different identities' for quantitative evaluation and user study. However, it does not specify explicit training/validation/test splits with percentages or counts for these datasets. |
| Hardware Specification | Yes | We train both Ωg and Ωc for 100 iterations with one Ge Force RTX 2080 Ti GPU. |
| Software Dependencies | No | The paper mentions using various tools and models like Control Net, Open Pose, SAM, DPT, ICON, and IP-Adapter, but does not provide specific version numbers for any of them. |
| Experiment Setup | Yes | During the geometry optimization stage, we set λP SL, λnorm, and λlap to 10,000, and set λnorm SDS to 1. During the texture optimization stage, we set λrecon and λtex SDS to 10,000 and 1, respectively. |