Main Module
The main entry point for the pytorch-segmentation-models-trainer CLI and core functionality.
Model Classes
This page is the API reference for all PyTorch Lightning model classes in pytorchsegmentationmodels_trainer.
Dataset Classes
This page is the API reference for all dataset classes in pytorchsegmentationmodels_trainer.
Loss Functions
This page is the API reference for all loss classes in pytorchsegmentationmodels_trainer.
Inference Processors
Inference processors orchestrate the full prediction pipeline for a trained model: reading a geospatial image, slicing it into tiles, running the model on each tile, merging the results back into a full-resolution output, and saving the inference to disk.
Custom Callbacks
Custom callbacks extend PyTorch Lightning's Callback and BasePredictionWriter interfaces to add visualisation, metrics reporting, loss normalisation, and polygonisation steps at specific points in the training or prediction loop.
Custom Optimizers
The project provides drop-in replacements for the standard PyTorch optimizers that add optional Gradient Centralization (GC) support. They are located in:
Mask Builder API
The Mask Builder system generates multi-channel training mask files from geospatial vector data (polygons) paired with raster images. For each input image it can produce up to six mask types, a CSV dataset manifest, and optionally bounding-box or polygon list files.