Skip to content
Gyanateet Dutta
Work·Tools·Academic·Figures·CV

Work

Systems

B3tt3r: 3D Reconstruction

Stereo pairs are the kind of problem I come back to: two pictures, and a surface that has to agree with both. I combined Mast3r and Spann3r for matching, depth, and a mesh.

Two photographs of a brick mill beside a small grey paper model of a building on a white table.
Fig. 1Photographs and a paper model, standing in for a stereo reconstruction. Generated illustration. Not a capture from the B3tt3r repository.
Two ink drawings of the same mill, seen from opposite sides, cycling through matched corners, a hatched depth, and a mesh.
Fig. 2A drawn turntable of one stereo pair: the same mill in two views, first with matched corners, then hatched for depth, then tied into a mesh. Drawn turntable of the method. Not a capture from the B3tt3r repository, and not a measured reconstruction.
Scope

A working integration of the two models; benchmarking against the source papers is future work.

Stack
3D Vision, Stereo Matching, Python, Deep Learning
Links
Code

Also

  • 2026CUDA Blackwell Labs
  • 2026Agentic structure-from-motion
  • 2026PCOS edge agent
  • 2024Dalton Mills VR Reconstruction
  • 2023AWS AI/ML Scholar
GitHub· ORCID· Google Scholar· LinkedIn· Hugging Face· X