The field of virtual try-on and avatar modelling is witnessing significant advancements, particularly in the areas of high-fidelity garment fitting and realistic human reconstruction. Innovations are being driven by the integration of advanced machine learning techniques such as Diffusion Transformers and Reinforcement Learning, which are enhancing the precision and flexibility of virtual try-on systems. Notably, there is a shift towards more user-friendly and adaptable models that can handle diverse scenarios and low-quality inputs, reducing the dependency on high-quality standing images. Additionally, the field is making strides in the separation of garments from the body in avatar models, enabling more realistic and editable representations. These developments are not only improving the visual quality of virtual try-ons but also expanding their practical applications in online shopping and human animation.

Noteworthy Papers:

• GGAvatar introduces a novel approach to garment-separated avatar reconstruction from monocular videos, achieving superior quality and efficiency.
• Try-On-Adapter proposes an outpainting paradigm for virtual try-on, offering flexible control and handling low-quality inputs effectively.
• FitDiT advances high-fidelity virtual try-on with enhanced garment perception, excelling in texture-aware maintenance and size-aware fitting.