How schillij95/thaumatoanakalyptor works
Project type and purpose
• This is a specialized full-stack data-processing and visualization pipeline (not a web app) that automates segmentation of papyrus sheets from CT-scan volumes. It combines:
• A Python “orchestrator” (ThaumatoAnakalyptor.py) and CLI/GUI frontend
• Deep-learning 3D instance segmentation (Mask3D)
• Graph‐based patch stitching (C++ solver)
• Mesh reconstruction and texturing
• Utilities for pre/post-processing (point clouds, rendering, etc.)
File structure to extract architectural modules
• Top level
• DockerfileThaumato – defines containerizing environment
• GUI/ – PyQt GUI code
• ThaumatoAnakalyptor.py – entry point, dispatches CLI/GUI tasks
• ThaumatoAnakalyptor/ – core Python package
• Vesuvius-Grandprize-Winner/ – ink detection model (external repo submodule)
• In ThaumatoAnakalyptor/:
• mask3d/ – 3D instance-segmentation neural net (PyTorch, configs, models, training scripts)
• graph_problem/ – C++ code and CUDA solver for sheet-patch optimization
• sheet_generation/ – C++ library and Python bindings for HDBSCAN clustering and sheet‐stitching
• Utilities – pointcloud/mesh converters, rendering, BP node deactivation, belief_propagation modules
• Data folders under mask3d (conf/, data/, models/, scripts/, third_party, trainer, utils) show a standard ML training/inference sub-architecture.
README for additional insights
• The README describes a multi-stage pipeline:
a. Volume preprocessing & grid generation
b. PointCloud extraction
c. 3D instance segmentation (Mask3D)
d. Patch graph construction & optimization (graph_problem)
e. Mesh reconstruction (Poisson)
f. Texture unwrapping (Volume Cartographer)
g. GUI for interactive use
Detailed capturing of essential elements
• Third-party modules (pointnet2, HDBSCAN) as sub-blocks within mask3d/sheet_generation
• Denote entry point “ThaumatoAnakalyptor.py” as the central orchestrator