Description
A 3D avatar of yourself usually means a studio and a fixed set of poses. UP2You, the official code for an ICLR 2026 paper, reconstructs a 3D person from an unconstrained photo collection with varied poses and backgrounds, fast and without per-person tuning.
The repo includes environment setup, pretrained models on Hugging Face and inference scripts for researchers and developers working on avatars or virtual try-on.
Casual photos:No fixed poses or backgrounds.
Tuning-free:No per-person training.
Fast:Efficient inference pipeline.
Open weights:Pretrained models to download.
The repo includes environment setup, pretrained models on Hugging Face and inference scripts for researchers and developers working on avatars or virtual try-on.
Features
Casual photos:No fixed poses or backgrounds.
Tuning-free:No per-person training.
Fast:Efficient inference pipeline.
Open weights:Pretrained models to download.
