README.md

metadata

title: GLPN Single-View Depth → Point Cloud → Mesh
emoji: 🧱
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 4.44.1
app_file: app.py
python_version: '3.10'
pinned: true
models:
  - vinvino02/glpn-nyu
tags:
  - depth-estimation
  - 3d-reconstruction
  - point-cloud
  - mesh
  - open3d
  - trimesh
  - transformers
  - gradio
  - single-view
  - computer-vision

Single-View Depth → Point Cloud → Mesh (GLPN + Open3D)

This Space predicts depth from a single RGB image using GLPN (NYU), back-projects to a point cloud, reconstructs a mesh (Ball Pivoting with Poisson fallback), and exports PLY and GLB.

How to use

1. Upload one RGB image (indoor works best).
2. Click Run.
3. Inspect Input vs Depth, Mesh Preview (if available), and the Interactive 3D (GLB) viewer.
4. Download artifacts from the Downloads panel.

Outputs

• input_vs_depth.png – side-by-side original and depth (meters, contrast-clipped)
• point_cloud.ply – colorized point cloud
• mesh.ply – triangle mesh
• mesh.glb – mesh for web viewers, Gradio Model3D, and downstream tools
• mesh_preview.png – headless snapshot (may be absent in some environments)

Notes

• OpenMP warnings fixed by setting safe thread env vars at the top of app.py.
• GPU is optional. The app auto-selects CUDA if available.
• If GLB doesn’t render: check Logs / Status. The exporter is skipped if the mesh is empty.
• For crisper meshes, try higher-texture images with clear geometry and lighting.