srforge 0.8.0

Super‑resolution research framework for PyTorch with a focus on simplicity and flexibility using config files.

SR FORGE

Super-Resolution Framework for Oriented Restoration and Guided Enhancement

---

SR FORGE (Super-Resolution Framework for Oriented Restoration & Guided Enhancement) is a unified, modular, and task-driven framework for training and evaluating deep learning models in the field of super-resolution.

Key Features

• Structured Workflow
  SR FORGE provides an organized approach to super resolution. Every step—from data loading to final evaluation—follows a clear, modular structure.

• Task-driven restoration
  Built-in utilities to help fine-tune models for specific tasks or objectives (e.g., OCR, remote sensing, medical imaging, etc.).

• Config-Driven Experiments
  Simple YAML/JSON configuration files let you customize your pipeline without modifying code directly.

• Flexible Model Plug-In
  Includes SISR baselines (FSRCNN, DSen2) and MISR models (RAMS, TR-MISR, MagNAt), plus a registry for custom architectures.

• Unified Metrics
  Evaluate your models with a suite of standard metrics (PSNR, SSIM, LPIPS) and straightforward logging.

• Visualization Tools
  Quickly visualize results (side-by-side comparisons, zoom-ins, or overlays) for interpretability and debugging.

Installation

1. Clone the Repository

   git clone https://github.com/your-username/sr-forge.git
   cd sr-forge
   

2. Install

   pip install -e .
   

Testing

SR FORGE uses pytest (industry standard for Python) for unit tests. Tests run to completion and provide a summary of passes/failures by default (similar to GoogleTest in C++).

Run transform tests

python -m pytest -q tests/transform/test_entry_transforms.py

Run dataset tests

python -m pytest -q tests/dataset

Run model tests

python -m pytest -q tests/models

Useful pytest configurations

• Verbose per-test output

  python -m pytest -v
  

• Full summary of passes/failures

  python -m pytest -rA
  

• Never stop early

  python -m pytest --maxfail=0
  

• Common "gtest-like" summary

  python -m pytest -v -rA --maxfail=0
  

Optional dependencies and skips

Some tests require optional dependencies (e.g., torch_geometric). These tests are automatically skipped if the dependency is missing, and pytest will report them as skipped in the summary.