CRUSH.md

CRUSH Guidelines for HF Demo

Build/Run Commands

• Setup: pip install -r requirements.txt
• Run Demo: python app.py
• Run Main: python main.py

Lint/Test Commands

• Lint: flake8 . or ruff check .
• Format: black . or ruff format .
• Type Check: mypy --ignore-missing-imports .
• Single File Test: python -m pytest tests/test_file.py::test_function_name -v

Code Style Guidelines

• Language: Python 3.12+, PyTorch-based
• Imports: Standard library, third-party (torch, torchvision, gradio), then local
• Formatting: Black-compatible, 88 char line limit
• Naming: snake_case for functions/variables, PascalCase for classes
• Types: Use type hints where possible (torch.Tensor, etc.)
• Docstrings: Not required for this demo app
• Error Handling: Use try/except for file I/O and model operations

Key Architecture Components (HF Demo)

• FocalCrossViTHybrid: Main hybrid model combining Focal Transformer + CrossViT
• FocalModulation: Adaptive spatial attention with multi-scale focal windows
• CrossAttention: Cross-attention mechanism between near/far focus images
• Gradio Interface: Web UI for image fusion demo

Important Conventions

• Models loaded from Hugging Face automatically
• Use torch.device('cuda' if torch.cuda.is_available() else 'cpu')
• Image size: 224x224 for model compatibility
• Follow Gradio patterns for UI components
• Include proper error messages for missing inputs

File Structure

• app.py: Main Gradio application with model definition and UI
• main.py: Simple entry point
• requirements.txt: Dependencies for HF Spaces

Test Commands

• No formal tests in this directory
• Test locally by running python app.py and uploading images