2 papers across 2 sessions
We propose a pipeline for detecting visual inconsistencies using visual correspondences based on disentangled diffusion model features
We present TIRE, a novel method for subject-driven 3D/4D generation that well preserves the identity.