feat: Set up comprehensive Python testing infrastructure with Poetry

[llbbl]

Add Python Testing Infrastructure with Poetry

Summary

This PR sets up a comprehensive testing infrastructure for the MAT (Mask-Aware Transformer) project using Poetry as the package manager and pytest as the testing framework. The setup provides a solid foundation for writing and running tests with proper coverage reporting and organization.

Changes Made

Package Management

• Migrated to Poetry: Created pyproject.toml with all dependencies from requirements.txt
• Lock file: Generated poetry.lock for reproducible builds
• Development dependencies: Added pytest, pytest-cov, and pytest-mock

Testing Configuration

• pytest settings: Configured in pyproject.toml with:
  • Test discovery patterns
  • Coverage reporting (term, HTML, XML)
  • 80% coverage threshold (configurable)
  • Custom markers: unit, integration, slow
  • Verbose output and strict mode

Directory Structure

tests/
├── __init__.py
├── conftest.py          # Shared fixtures
├── test_infrastructure_validation.py
├── unit/
│   └── __init__.py
└── integration/
    └── __init__.py

Testing Fixtures (in conftest.py)

• temp_dir: Temporary directory management
• mock_config: Configuration dictionary for testing
• sample_image, sample_mask: NumPy arrays for image testing
• sample_tensor_image, sample_tensor_mask: PyTorch tensors
• mock_model: Simple PyTorch model for testing
• mock_dataset_path: Creates mock dataset structure
• device: CPU/GPU device selection
• reset_random_seeds: Ensures reproducible tests

Additional Setup

• Updated .gitignore with testing artifacts and Poetry files
• Created validation tests to verify the infrastructure works
• Configured Poetry scripts for easy test execution

How to Use

1. Install dependencies:

   poetry install

2. Run tests:

   poetry run test
   # or
   poetry run tests

3. Run specific test markers:

   poetry run pytest -m unit        # Run only unit tests
   poetry run pytest -m integration # Run only integration tests
   poetry run pytest -m "not slow"  # Skip slow tests

4. View coverage reports:

   • Terminal: Automatically shown after test run
   • HTML: Open htmlcov/index.html in browser
   • XML: Available at coverage.xml for CI/CD integration

Notes

• The coverage threshold is initially set to 80% but can be adjusted in pyproject.toml
• The infrastructure validation tests all pass, confirming proper setup
• Poetry lock file is committed for reproducible environments
• No actual unit tests for the codebase were written - only infrastructure setup

Next Steps

Developers can now immediately start writing tests for:

• Model components (networks/)
• Loss functions (losses/)
• Metrics (metrics/)
• Data processing (datasets/)
• Training loops (training/)
• Utilities (torch_utils/, dnnlib/)