How iamhharish/polyatomic-nodules-detection works
This repository is a seabed-nodule image-processing and detection prototype. Its image workflow preprocesses raw images, then provides separate unsupervised and supervised training scripts; detection independently applies classical edge-and-contour processing to a processed image. The sampled detection code does not load either trained model, so the graph does not imply a model-to-detection connection. Training inputs and outputs are inferred from directory and filename references in the scripts; no wiring for the YOLO files is established by the sampled source.
Main entry points: preprocess_images.py Image preprocessing · train_unsupervised_dbn.py Unsupervised DBN · train_supervised_dbn.py Supervised classifier · detect_objects.py Contour detection
Components 10
Image preparation
- Image preprocessing
preprocess_images.py - Processed images
Model training
- Unsupervised DBN
train_unsupervised_dbn.py - DBN feature weights
- Supervised classifier
train_supervised_dbn.py - Classifier weights
Image detection
- Contour detection
detect_objects.py - Annotated image
Other components
- Raw images
- Training images
Connections 8
- Raw images → Image preprocessing: reads
- Image preprocessing → Processed images: writes
- Processed images → Unsupervised DBN: reads
- Unsupervised DBN → DBN feature weights: saves weights
- Training images → Supervised classifier: reads
- Supervised classifier → Classifier weights: saves weights
- Processed images → Contour detection: reads
- Contour detection → Annotated image: writes