Cascaded Diffusion Models for Virtual Try-On: Improving Control and Resolution

Authors: Guangyuan Li, Yongkang Wang, Junsheng Luan, Lei Zhao, Wei Xing, Huaizhong Lin, Binkai Ou

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results demonstrate that our method outperforms previous approaches in preserving garment details and generating authentic virtual try-on images, both qualitatively and quantitatively.
Researcher Affiliation Collaboration 1College of Computer Science and Technology, Zhejiang University 2 Innovation Research & Development, Board Ware Information System Limited
Pseudocode No The paper describes the model architecture and methodology in detail using text and diagrams (e.g., Fig. 2), but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statement about releasing source code, nor does it provide any links to code repositories.
Open Datasets Yes We employ two publicly available datasets, Dress Code (Morelli et al. 2022) and VITON-HD (Choi et al. 2021), to evaluate the virtual try-on task. Both datasets consist of paired images of garments and their corresponding human models wearing the garments.
Dataset Splits No The paper mentions testing experiments are conducted under 'paired' and 'unpaired' settings but does not provide specific numerical splits (e.g., percentages or counts) for training, validation, or testing datasets.
Hardware Specification Yes Specifically, MC-DM is conducted using two NVIDIA A6000 (48GB) GPUs... SR-DM is trained on two NVIDIA A100 (80GB) GPUs
Software Dependencies No The paper mentions using the Adam W optimizer but does not specify version numbers for any programming languages, libraries, or frameworks (e.g., Python, PyTorch, TensorFlow).
Experiment Setup Yes Specifically, MC-DM is conducted using two NVIDIA A6000 (48GB) GPUs with image resolutions of 512 384. We use the Adam W optimizer with a learning rate set to 2e-5. SR-DM is trained on two NVIDIA A100 (80GB) GPUs, employing the Adam W optimizer with a learning rate of 5e-5. In MC-DM, we use Paintby-Example (Yang et al. 2023) as the frozen pre-trained diffusion model. In SR-DM, we use IRControl Net (Lin et al. 2024) as the frozen pre-trained diffusion model.