3D Vision & Generative AI

Digit3D: 3D Multimodal MNIST Benchmark

Transforming 2D MNIST into rich 3D geometries (watertight meshes, sparse SDFs, and 6D normal-oriented point clouds) with continuous flow matching generative pipelines.

Khoi DO
Conquer3D Research & Open-Source Geometry Engine
Code Conquer3D Engine Overview Unconditional Generation Class-Conditional Generation Image-Conditional Generation Arbitrary-Resolution Generation Sparse Voxel Reconstruction Image-Conditional Mesh Generation PointCloud-Conditional Mesh Generation Limitations
Experiment 3

2D Image → 3D Point Cloud Generation

Single-view 3D reconstruction and generative shape synthesis. A vision encoder maps raw 28 × 28 grayscale images through GroupNorm feature extractors to steer Mean Flow generation.

0 Sample #004
2D Input Digit 0 2D Input
1 Sample #000
2D Input Digit 1 2D Input
2 Sample #002
2D Input Digit 2 2D Input
3 Sample #000
2D Input Digit 3 2D Input
4 Sample #005
2D Input Digit 4 2D Input
5 Sample #008
2D Input Digit 5 2D Input
6 Sample #001
2D Input Digit 6 2D Input
7 Sample #000
2D Input Digit 7 2D Input
8 Sample #000
2D Input Digit 8 2D Input
9 Sample #016
2D Input Digit 9 2D Input

Drag to rotate 3D point cloud • Scroll to zoom • Shaded by continuous surface normal vectors