GAE: A Geometry-Native Latent Space for 3D-Consistent World Generation
Loading video
Loading videoTencent ARC GAE compresses 3,072-channel geometry features into a 128-channel latent that decodes into RGB, depth, camera trajectories and point clouds from one generated state, halving camera-trajectory error and cutting FVD by up to 23%.
Category: research
Author: @HuggingPapers
Date: 2026-09-23T00:00:00
Duration: 51.396s
Reference: https://arxiv.org/abs/2609.24981
