Related Experiment Video
Updated: Jun 26, 2025
09:17
Using a Virtual Store As a Research Tool to Investigate Consumer In-store Behavior
Published on: July 24, 2017
11.3K
A novel garment transfer method supervised by distilled knowledge of virtual try-on model
Naiyu Fang1, Lemiao Qiu1, Shuyou Zhang1
1State Key Laboratory of Fluid Power & Mechatronic Systems, Zhejiang University, Hangzhou, 310027, China.
Summary
This study introduces a novel garment transfer method using knowledge distillation from virtual try-on. It achieves state-of-the-art results by improving robustness and realism in transferring clothing between images.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Image Processing
Background:
- Garment transfer aims to apply clothing from a source image to a target person.
- Existing methods struggle with robustness and completeness due to lack of ground truth data.
- Virtual try-on technologies show promise but differ in application.
Purpose of the Study:
- To develop a robust and realistic garment transfer method.
- To overcome limitations of self-supervised learning in garment transfer.
- To leverage virtual try-on successes for garment transfer.
Main Methods:
- Knowledge distillation from virtual try-on models to supervise garment transfer training.
- A pipeline involving garment transfer parsing, progressive flow warping, and inpainting.
- An arm regrowth task to enhance realism by inferring exposed skin.
Main Results:
- The proposed method achieves state-of-the-art performance in cross-person garment transfer.
- Demonstrated superior robustness and realism compared to existing methods.
- Successfully integrates parsing, warping, and inpainting for effective transfer.
Conclusions:
- Knowledge distillation from virtual try-on is an effective strategy for garment transfer.
- The proposed pipeline addresses key challenges in garment transfer, including robustness and realism.
- This approach offers significant potential for commercial applications in fashion and e-commerce.
