Skip to content

README.md

Latest commit

 

History

History
150 lines (110 loc) · 3.32 KB

README.md

File metadata and controls

150 lines (110 loc) · 3.32 KB

CABiNet Test Suite

This directory contains the test suite for the CABiNet semantic segmentation project.

Test Structure

tests/
├── __init__.py
├── conftest.py              # Shared fixtures and configuration
├── unit/                    # Unit tests for individual components
│   ├── test_models.py       # Model architecture tests
│   ├── test_loss.py         # Loss function tests
│   └── test_transforms.py   # Data augmentation tests
├── integration/             # Integration tests
│   └── test_training_pipeline.py  # End-to-end training tests
└── fixtures/                # Test data and fixtures

Running Tests

Run All Tests

pytest tests/

Run Specific Test Category

# Unit tests only
pytest tests/unit/

# Integration tests only
pytest tests/integration/

# Specific test file
pytest tests/unit/test_models.py

# Specific test function
pytest tests/unit/test_models.py::TestCABiNetModel::test_cabinet_forward_shape

Run with Coverage

pytest tests/ --cov=src --cov-report=html

This generates an HTML coverage report in htmlcov/index.html.

Run in Parallel

pytest tests/ -n auto

Uses all available CPU cores for faster testing.

Run with Verbose Output

pytest tests/ -v

Test Markers

Tests can be marked for selective execution:

@pytest.mark.slow  # Slow tests
@pytest.mark.gpu   # Requires GPU
@pytest.mark.integration  # Integration tests

Run specific markers:

pytest tests/ -m "not slow"  # Skip slow tests
pytest tests/ -m gpu        # Run only GPU tests

Writing Tests

Unit Test Example

def test_model_forward():
    """Test model forward pass."""
    model = CABiNet(n_classes=19, mode="large")
    x = torch.randn(1, 3, 512, 512)

    out, out16 = model(x)

    assert out.shape == (1, 19, 512, 512)
    assert not torch.isnan(out).any()

Using Fixtures

def test_with_fixture(sample_image, num_classes):
    """Test using shared fixtures."""
    model = CABiNet(n_classes=num_classes, mode="large")
    out, _ = model(sample_image)

    assert out.shape[0] == sample_image.shape[0]

Parametrized Tests

@pytest.mark.parametrize("mode", ["large", "small"])
def test_multiple_modes(mode, num_classes):
    """Test with different model modes."""
    model = CABiNet(n_classes=num_classes, mode=mode)
    # ... test code

Continuous Integration

Tests are automatically run on:

  • Every push to main/develop branches
  • Every pull request
  • Multiple Python versions (3.8, 3.9, 3.10, 3.11)
  • Multiple OS (Ubuntu, macOS)

See .github/workflows/ci.yml for CI configuration.

Test Coverage Goals

  • Unit Tests: >80% coverage
  • Integration Tests: Critical paths covered
  • Model Tests: All forward/backward passes tested
  • Loss Tests: All loss functions validated

Performance Tests

Use the performance profiler for benchmarking:

from src.utils.profiler import PerformanceProfiler

profiler = PerformanceProfiler(model)
results = profiler.run_full_benchmark()

Contributing

When adding new features:

  1. Write tests first (TDD)
  2. Ensure tests pass locally
  3. Check coverage: pytest --cov=src
  4. Run pre-commit hooks: pre-commit run --all-files

See CONTRIBUTING.md for more details.