vascx-simplify 0.1.2

A PyTorch library for vessel and fundus image analysis (simplified rewrite)

vascx_simplify

⚠️ IMPORTANT: This project is NOT original work.

This is a simplified rewrite of the original rtnls_vascx_models project by Eyened. All core algorithms, preprocessing logic, and model architectures are derived from that original work.

AI Usage Disclaimer

This project was developed with significant assistance from AI tools (GitHub Copilot, ChatGPT, Claude). AI was used extensively for:

• Code organization and refactoring
• Documentation and comments
• Testing infrastructure
• Packaging and build configuration
• Docker setup
• CI/CD pipeline

The underlying algorithms and models are from the original work referenced above.

---

A PyTorch library for vessel and fundus image analysis, providing GPU-accelerated preprocessing, contrast enhancement, and inference utilities for medical imaging tasks.

Features

• GPU-Accelerated Preprocessing: Fast fundus image contrast enhancement with mixed precision (FP16/FP32) support
• Sliding Window Inference: Efficient inference for large images with MONAI-compatible implementation
• Ensemble Models: Support for segmentation, classification, regression, and heatmap-based models
• HuggingFace Integration: Easy model loading from HuggingFace Hub
• Flexible Transformations: Built-in image preprocessing and normalization

Installation

From PyPI (once published)

pip install vascx_simplify

From source

git clone https://github.com/kapong/vascx_simplify.git
cd vascx_simplify
pip install -e .

With uv (recommended)

uv pip install vascx_simplify

Quick Start

import torch
from vascx_simplify import VASCXTransform, EnsembleSegmentation, from_huggingface

# Download model from HuggingFace
model_path = from_huggingface("username/model-name:model.pt")

# Setup preprocessing
transform = VASCXTransform(size=1024, have_ce=True, device='cuda')

# Load model
model = EnsembleSegmentation(
    fpath=model_path,
    transforms=transform,
    device='cuda'
)

# Process an image
import numpy as np
from PIL import Image

image = Image.open("fundus_image.jpg")
prediction = model.predict(image)

Usage Examples

Preprocessing Only

from vascx_simplify import FundusContrastEnhance
import torch

# Initialize enhancer
enhancer = FundusContrastEnhance(
    square_size=1024,
    sigma_fraction=0.05,
    contrast_factor=4,
    use_fp16=True  # Use FP16 for faster processing on GPU
)

# Process image (torch.Tensor [C, H, W])
rgb, enhanced, bounds = enhancer(image)

Sliding Window Inference

from vascx_simplify import sliding_window_inference
import torch

# Your model
class MyModel(torch.nn.Module):
    def forward(self, x):
        # Your implementation
        return x

model = MyModel()
inputs = torch.randn(1, 3, 2048, 2048).cuda()

# Perform inference
output = sliding_window_inference(
    inputs=inputs,
    roi_size=(512, 512),
    sw_batch_size=4,
    predictor=model,
    overlap=0.5,
    mode='gaussian'
)

Development

Setup Development Environment

# Clone the repository
git clone https://github.com/kapong/vascx_simplify.git
cd vascx_simplify

# Install with development dependencies
pip install -e ".[dev,test]"

Running Tests

Tests have been removed from this simplified version.

Code Quality

# Format code
black src/

# Sort imports
isort src/

# Lint code
flake8 src/

# Type checking
mypy src/

Requirements

• Python >= 3.8
• PyTorch >= 1.10.0
• kornia >= 0.6.0
• scikit-learn >= 1.0.0
• scipy >= 1.7.0
• numpy >= 1.21.0
• huggingface-hub >= 0.10.0

Contributing

This is a simplified rewrite for personal use. For contributions to the original work, please visit rtnls_vascx_models.

License

MIT License - see LICENSE file for details.

Citation

If you use this library in your research, please cite the original work:

@software{vascx_simplify,
  title = {vascx_simplify: Simplified rewrite of rtnls_vascx_models},
  author = {kapong},
  year = {2025},
  url = {https://github.com/kapong/vascx_simplify},
  note = {Simplified rewrite of https://github.com/Eyened/rtnls_vascx_models}
}

Acknowledgments

Original Work: rtnls_vascx_models by Eyened - All core algorithms and models are from this original work.

AI Assistance: This rewrite was created with extensive assistance from AI tools (GitHub Copilot, ChatGPT, Claude) for code organization, documentation, and packaging.