← All threads

Sparse 3D Generation

Methods for generating high-resolution 3D geometry from single images using sparse flow matching or diffusion.

4 papers

Where this stands

The written synthesis of this thread is for subscribers. Subscribe.

How this thread developed

  1. Oct 2026 · Alibaba Group

    Sparse flow matching generates 3D shapes at ultra-high voxel resolution

    Generates ultra-high resolution 3D shapes from a single image using sparse flow matching, scalable to 4096³ voxels.

  2. 1 further paper

    Oct 2026 · Technical University of Denmark, Pioneer Center for Artificial Intelligence

    Aligning partial point clouds to frozen global 3D encoders without labels

    Proposes an alignment method for partial point clouds to frozen global 3D encoders, improving zero-shot generalization in sparse 3D tasks

  3. Oct 2026 · Preferred Networks, Inc., The University of Tokyo

    Single generator reconstructs 3D scenes at any scale without per-scene optimization

    Adds a method for scalable 3D mesh reconstruction from posed images using a single generative model without per-scene optimization.

3 of 4 papers shown