PIF2024 - Generating realistic and dynamic avatars in unconstrained environments (PID2023-151184OB-I00)

The overarching aim of this project is to advance the field of 3D human digitization, interaction modeling, and related technologies. Through cutting-edge research and innovation, we seek to contribute to the development of highly realistic and personalized avatars with applications spanning virtual reality, entertainment, e-commerce, healthcare, and fitness monitoring. Particularly, in this project you will work on developing novel algorithms that enable high-fidelity 3D human digitization from RGB images in non-controlled settings. Address challenges related to noisy observations, non-calibrated cameras, and diverse body poses to improve robustness in real-world scenarios. To this end, we will explore some novel ideas in neural rendering, Gaussian splatting and diffusion models.