witchcraftcode/crackdetection

How witchcraftcode/crackdetection works

This repository provides a concrete-crack image classifier and inspection toolkit. Its main workflow trains a binary classifier from labeled images, saves a model, then serves predictions through an image-upload API or runs them from scripts. The README also documents video, explanation, segmentation, edge export, drone simulation, architecture benchmarking, and metrics evaluation. The graph includes those distinct capabilities; connections for unsampled script internals rely on README documentation or SOURCE INDEX references, not inspected implementations.

Main entry points: inference.py Image Inference · train_classifier.py Classifier Training · segment_cracks.py Attention Masks · model.py Cached Model Helper

Components 18

Model Training

Classifier Training scripts/train_classifier.py
Classifier Architectures src/architectures.py
Trained Classifier

Prediction Interfaces

Prediction API app.py
Cached Model Helper model.py
Image Inference src/inference.py

Inspection Tools

Video Inference scripts/video_inference.py
Image Explanations scripts/explain_image.py
Attention Masks scripts/segment_cracks.py
Segmenter Training scripts/train_segmentation.py
U-Net Architecture src/segmentation.py
Drone Simulation scripts/drone_inspection_sim.py

Evaluation and Deployment

Metrics Evaluation scripts/evaluate_metrics.py
Architecture Benchmark scripts/benchmark_architectures.py
Edge Model Export scripts/export_edge_model.py

Other components

Labeled Images
API Client
Image User

Connections 23

Labeled Images → Classifier Training: trains on README.md
Classifier Training → Classifier Architectures: selects architecture scripts/train_classifier.py
Classifier Training → Trained Classifier: saves model scripts/train_classifier.py
API Client → Prediction API: uploads image README.md
Image User → Image Inference: requests prediction README.md
Prediction API → Cached Model Helper: calls prediction app.py
Cached Model Helper → Image Inference: loads and predicts model.py
Trained Classifier → Cached Model Helper: loaded by model.py
Trained Classifier → Video Inference: loads for frames scripts/video_inference.py
Video Inference → Image Inference: loads model scripts/video_inference.py
Trained Classifier → Image Explanations: explained by README.md
Image Explanations → Image Inference: loads and preprocesses scripts/explain_image.py
Trained Classifier → Attention Masks: generates attention mask README.md
Attention Masks → Image Explanations: uses Grad-CAM scripts/segment_cracks.py
Attention Masks → Image Inference: loads and preprocesses scripts/segment_cracks.py
Labeled Images → Segmenter Training: provides paired images and masks README.md
Segmenter Training → U-Net Architecture: builds U-Net scripts/train_segmentation.py
Trained Classifier → Drone Simulation: used for inspection README.md
Drone Simulation → Image Inference: uses prediction scripts/drone_inspection_sim.py
Trained Classifier → Metrics Evaluation: evaluated by README.md
Metrics Evaluation → Image Inference: uses preprocessing scripts/evaluate_metrics.py
Classifier Architectures → Architecture Benchmark: benchmarked by scripts/benchmark_architectures.py
Trained Classifier → Edge Model Export: exports for edge README.md

Drawn by GitDiagram on November 11, 2026 from the repository's file tree, README and source files. Read as Markdown ·