Alibaba's TaoFlowForge: Image-to-3D That Keeps Thin Parts, Chamfer 0.0655 vs 0.1553
jiqizhixin · x · 2026-11-20
Alibaba's Taobao3D team presents TaoFlowForge in an arXiv preprint, tackling a production pain point of image-to-3D: dense, irregular triangulation makes meshes hard to edit, while simplification erases thin structures.
Approach: The model decouples vertex placement from connectivity prediction. Cascaded flow matching first builds coarse structure, then progressively refines vertices; a final stage estimates vertex connectivity and predicts normals. Geometry-aware losses encourage structurally important vertices to survive.
Results: Under single-image conditioning on Toys4K, it reports a Chamfer distance of 0.0655, versus 0.1553 for LATO.2 and 0.3264 for EdgeRunner, while producing artist-editable meshes.
More from Multimodal
- Designer: Fable nails trailer hitch models, "this could be an entire product" — tomjohndesign · 2026-11-20
- Mixing two --sref style codes yields moody fog-forest Midjourney figures in blue — sergeantsref · 2026-11-20
- Pixel-art Gaussian splat effects are getting fun, made in astra 6.2 — nptacek · 2026-11-20
- AI-generated Manny Calavera from Grim Fandango nails the iconic look — evilrabbit_ · 2026-11-20
- Stickman fighting demos generated by swarms of Gemini 4 Neon in pure JS — cgarciae88 · 2026-11-20
- 1-bit pixel-art trailer made with Pixel Rabbit and OpenAI tools — CommentBig3066 · 2026-11-20