Tencent ARC's GAE: Geometry-Native Latent Space Halves Camera Error in World Generation
yshan2u · x · 2026-09-23
Tencent ARC released GAE, a geometry-native latent space for 3D-consistent world generation.
- Compresses 3,072-channel geometry features into a compact 128-channel latent
- A single generated latent decodes into RGB, depth, camera trajectories, and point clouds
- Halves camera-trajectory error and cuts FVD by up to 23% on standard benchmarks
By baking geometry directly into the latent space instead of stitching modalities afterward, GAE is an architecture worth watching in the world-model space.
More from Research
- TLAPS-Bench launches in alpha to test whether AI agents can formally prove system correctness — tianyin_xu · 2026-09-23
- ImIR replaces text prompts with image instructions for all-in-one restoration — Süleyman Aslan · 2026-09-23
- Quanta Explains How Pricing Algorithms Can Drive Up Prices Without Collusion — burny_tech · 2026-09-23
- CodeMidas: Turning Raw Source Code into Executable RL Environments for Coding Agents — burny_tech · 2026-09-23
- Researchers pine for pre-2000s methodological papers with no mathiness or defensive fluff — PMinervini · 2026-09-23
- Block-triangular joint drifting enables one-step generative surrogate models for stochastic trajectories — chaumian · 2026-09-23