Reproduce “MFP-VTON: Enhancing Mask-Free Person-to-Person Virtual Try-On via Diffusion Transformer” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Le Shen, Yanting Kang, Rong Huang, Zhijie Wang
Researchers from Donghua University developed MFP-VTON, a mask-free framework for person-to-person virtual try-on utilizing a fine-tuned Diffusion Transformer. The approach generates high-fidelity try-on images, achieving the lowest FID of 10.3465 and CLIP-FID of 2.2885 in P2P tasks compared to adapted state-of-the-art methods.
Key insights
- High-quality synthetic paired data, generated by leveraging existing G2P models, can effectively mitigate data scarcity for training complex person-to-person virtual try-on models.
- Directly feeding full target person images to a diffusion transformer without explicit masking preserves crucial foreground details and maintains pose consistency, yielding more natural try-on results.
- Supervising the attention mechanism with a targeted loss function can effectively direct a large foundation model's focus to relevant input regions, optimizing performance for intricate cross-image transfer tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.