Leffa is a unified framework for controllable person image generation that enables precise manipulation of both appearance (i.e., virtual try-on) and pose (i.e., pose transfer).
Controllable person image generation aims to generate a person image conditioned on reference images, allowing precise control over the person’s appearance or pose. However, prior methods often distort fine-grained textural details from the reference image, despite achieving high overall image quality. We attribute these distortions to inadequate attention to corresponding regions in the reference image. To address this, we thereby propose learning flow fields in attention (Leffa), which explicitly guides the target query to attend to the correct reference key in the attention layer during training. Specifically, it is realized via a regularization loss on top of the attention map within a diffusion-based baseline. Our extensive experiments show that Leffa achieves state-of-the-art performance in controlling appearance (virtual try-on) and pose (pose transfer), significantly reducing fine-grained detail distortion while maintaining high image quality. Additionally, we show that our loss is model-agnostic and can be used to improve the performance of other diffusion models.
An overview of our Leffa training pipeline for controllable person image generation. The left is our diffusion-based baseline; the right is our Leffa loss. Note that Isrc and Itgt are the same image during training.
Qualitative visual results comparison with other methods. The input person image for the pose transfer is generated using our method in the virtual try-on. The visualization results demonstrate that our method not only generates high-quality images but also greatly reduces the distortion of fine-grained details.
Create a conda environment and install requirements:
conda create -n leffa python==3.10
conda activate leffa
cd Leffa
pip install -r requirements.txtRun locally:
python app.pyWe use this code for metric evaluation.
Our code is based on Diffusers and Transformers. We use SCHP and DensePose to generate masks and densepose in our Demo. We also referred to the code of IDM-VTON and CatVTON.
If you find our work helpful or inspiring, please feel free to cite it.
@article{zhou2024learning,
title={Learning Flow Fields in Attention for Controllable Person Image Generation},
author={Zhou, Zijian and Liu, Shikun and Han, Xiao and Liu, Haozhe and Ng, Kam Woh and Xie, Tian and Cong, Yuren and Li, Hang and Xu, Mengmeng and Pérez-Rúa, Juan-Manuel and Patel, Aditya and Xiang, Tao and Shi, Miaojing and He, Sen},
journal={arXiv preprint arXiv:2412.08486},
year={2024},
}