How jordan-pierce/coralnet-toolbox works
Project type and purpose
• This is a full-stack, desktop-based annotation and ML toolkit built in Python using PyQt for the GUI.
• Purpose: Provide an end-to-end pipeline for coral reef image/video annotation and analysis, integrating manual tools, AI-assisted models (SAM, YOLO family, AutoDistill, etc.), dataset management, model training, inference, and export to multiple formats (CoralNet, Viscore, GeoJSON, TagLab).
File structure
• Top-level entry point: coralnet_toolbox/main.py
• GUI layer: Qt windows, widgets, and tools under coralnet_toolbox/ (e.g., QtMainWindow.py, QtImageWindow.py, Annotations/…, Tools/…, Explorer/…)
• IO layer: Import/export handlers under IO/ (QtImport*.py, QtExport*.py, Open/Save project)
• Core utilities: Common/, utilities.py
• Raster management: Rasters/
• Machine-Learning modules: grouped under MachineLearning/ with subpackages for BatchInference, TrainModel, EvaluateModel, DeployModel, AutoDistill, SAM, SeeAnything, Tile, etc.
• Results processing: Results/
• Scripts, notebooks, docs and CI pipelines are auxiliary and support development/testing.
README
• Confirms architecture: “From data collection to deployment,” lists stages: Data Input → Annotation & Labeling → Machine Learning → Analysis & Export
• Enumerates core components: Manual tools, AI models (SAM, YOLO, OWLViT…), dataset & project management, training/tuning, batch/video inference, export.
Elements
• Rectangles for components
Final advice
• Shared utilities (Common, RasterManager) as a central library used by both GUI and ML.