Tencent ARC's GAE: geometry-native latent space cuts FVD by up to 23% in world generation
TencentARC · hf · 2026-09-23
Tencent ARC presents GAE (geometry-native autoencoder), arguing inconsistent 3D generation is a representation problem, not just modeling: generators evolve appearance-centric latents while perception models recover geometry in cross-view-aware space. GAE reparameterizes a geometry foundation model's features into a compact latent jointly decodable to appearance, depth, cameras, and point maps; a standard conditional flow then supports diverse generation tasks.
In controlled comparisons holding the generator and training fixed, swapping in the GAE latent improves both visual quality and independently measured 3D coherence: FVD falls 12.7% and 23.1% on RealEstate10K and DL3DV, and camera-trajectory error halves on RealEstate10K — showing the latent space is central to geometry-consistent generation and a shared perception-generation interface.
More from Multimodal
- Pooling-aware training cuts retrieval vectors 4x while boosting nDCG to 97.8% of unpooled — lateinteraction · 2026-11-11
- Generating accurate 3D building models from 150 planning PDFs with new Gemini model — fofrAI · 2026-11-11
- Stanford's DyCraft Uses Agentic Decomposition to Cut Physical Errors in Long-Horizon 4D Scene Generation — StanfordUniversity · 2026-11-11
- Midjourney SREF pick: --2523103199 for loose pencil-sketch portraits — tisch_eins · 2026-11-11
- H3 Image Tips: 16MP via ComfyUI Workflow Plus Detailed Prompts Yields Best Results — gatortux · 2026-11-11
- Gemini-generated endless loop of quantum cryostats self-assembling — hapticdata · 2026-11-11