FashionTailor: Controllable Clothing Editing for Human Images with Appearance Preserving
Authors: Jie Hou, Jianghong Ma, Xiangyu Mu, Haijun Zhang, Zhao Zhang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments validate the effectiveness of our meth-od for editing part-level human images in Structure Fashion dataset and real-scenarios. Category Method Publication FID KID LPIPS SSIM PSNR CLIP-S D-CLIP... Table 1: Quantitative comparison of our Fashion Tailor with the state-of-the-art text-to-image methods based on references. Table 2: Quantitative results for ablation studies. |
| Researcher Affiliation | Academia | 1Department of Computer Science, Harbin Institute of Technology, Shenzhen 518055, China 2Department of Computer Science, Hefei University of Technology, Hefei 230009, China EMAIL, EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology using textual explanations and diagrams (Figure 3, Figure 4, Figure 5), but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code, nor does it include any links to a code repository. |
| Open Datasets | Yes | Third, to evaluate the effectiveness of Fashion Tailor, we collected a large-scale clothing dataset, Structure Fashion, which includes thousands of fashion items with hundreds of combinations of neckline types, sleeve types, top lengths, and bottom lengths. Ultimately, we created an image-instruction-image triplet dataset of approximately six million scales. Please refer to the Supplementary Material for a detailed introduction to the motivation and dataset. To advance image editing techniques in fashion, we collected the first clothing structure dataset, as shown in Figure 6. Please refer to the Supplementary Material for more details. |
| Dataset Splits | No | Due to computational resource limitations, we did not use all available data for model training. Please refer to the Supplementary Material for more details. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running its experiments. |
| Software Dependencies | Yes | In this study, we utilized pretrained IP2P (Brooks, Holynski, and Efros 2023) and Stable Diffusion (Rombach et al. 2022) v1.5 to initialize the weights of the denoising UNet and Clothing Net, respectively. |
| Experiment Setup | Yes | For classifier-free guidance, we set SI and Sc to 1.5 and 7.5, respectively, and drop the conditions Is and c with a probability of 5%. |