This repository contains an implementation for performing 3D animal (quadruped) reconstruction from a monocular image or video. The system adapts the pose (limb positions) and shape (animal type/height/weight) parameters for the SMAL deformable quadruped model, as well as camera parameters until the projected SMAL model aligns with 2D keypoints and silhouette segmentations extracted from the input frame(s).
Hello, May I ask what the unity prior stands for and why you only optimize 20 betas? Thanks for your work!!