Obj Scanning and STL Breakdown Toolkit

A free workflow to reconstruct physical objects from sparse images into printable STL files for hobbyists.

Motivation

This project is driven by a desire to support close friends who are entering the world of hobbyist 3D printing. The goal is to provide a free, accessible tool that can:

  1. Reconstruct physical objects from sparse image collections or short video clips.
  2. Algorithmically “break down” these reconstructions into clean, printable STL files, bridging the gap between photogrammetry and slicer-ready assets.

Pipeline Planning

The development is split into three main research tracks to identify the most robust workflow for consumer hardware.

1. Reconstruction from Sparse Data

  • Meta MapAnything: Investigating the use of MapAnything for handling sparse image inputs.
  • Or, Mono-Depth Guided SLAM: to generate dense pointclouds from continuous video feeds if available.

2. Masking & Meshing

  • Per-Frame Logic: Background/foreground or text guided segmentation using SAM (Segment Anything Model) to isolate the object of interest before reconstruction.
  • Part Segmentation: Leveraging SAM3 language alignment to segment specific parts of an object (e.g., “scan just the handle”).

3. Feed-Forward Evaluation (Experimental)

  • PolyGen-style Architecture: Beyond traditional optimization (photogrammetry), I aim to evaluate single-pass or partial feed-forward approaches (like PolyGen) to generate mesh topology directly. This could offer cleaner, more “CAD-like” meshes compared to the noisy surfaces typical of Poisson reconstruction.