angelcv 0.1.7

Train and inference for Computer Vision models made easy.

AngelCV

AngelCV is an open-source, commercially-friendly computer vision library designed for ease of use, power, and extensibility.

AngelCV is a project by Angel Protection System, a company at the forefront of safeguarding schools, hospitals, and other vital community spaces. They specialize in intelligent security and surveillance systems, including cutting-edge firearm detection technology that provides critical, real-time information to 911 and first responders, playing a vital role in saving lives.

Our mission is to provide cutting-edge deep learning models and tools that you can seamlessly integrate into your projects, whether for research, personal use, or commercial applications. All our code and pre-trained models are under the Apache 2.0 License, giving you the freedom to innovate without restrictive licensing.

A note on our open-source commitment: Angel Protection System initially developed AngelCV to enhance its advanced computer vision capabilities for security applications. We are excited to share it with the open-source community to foster innovation and allow everyone to benefit from and contribute to its development.

✨ Why AngelCV?

• Open & Free for Commercial Use: Build your next big thing without worrying about licensing fees or restrictions. Our Apache 2.0 license covers both the library and our provided pre-trained models.
• State-of-the-Art Models: We start with robust implementations like YOLOv10 for object detection and plan to expand to other vision tasks (classification, segmentation, oriented bounding boxes) and model architectures.
• Developer-Friendly Interface: A clean, intuitive API (see ObjectDetectionModel and InferenceResult) makes common tasks like training, inference, and evaluation straightforward.
• Flexible Configuration: Easily customize model architectures, training parameters, and datasets using YAML-based configuration files.
• Community Driven (Future): We aim to build a community around AngelCV.

🚀 Getting Started

Installation

AngelCV will be available on PyPI. You can install it using pip:

pip install angelcv

Make sure you have PyTorch installed, as it's a primary dependency. You can find PyTorch installation instructions at pytorch.org.

Quick Start: Object Detection

Here's a simple example of how to load a pre-trained YOLOv10 model and perform inference on an image:

from angelcv import ObjectDetectionModel

# Load a pre-trained YOLOv10n model (will download if not found locally)
# You can also specify a path to a local .ckpt or .pt file,
# or a .yaml configuration file to initialize a new model.
model = ObjectDetectionModel("yolov10n.ckpt")

# Perform inference on an image
# Source can be a file path, URL, PIL image, torch.Tensor, or numpy array.
results = model.predict("path/to/your/image.jpg")

# Process and display results
for result in results:
    print(f"Found {len(result.boxes.xyxy)} objects.")
    # Access bounding boxes (various formats available, e.g., result.boxes.xyxy_norm)
    # Access confidences: result.boxes.confidences
    # Access class IDs: result.boxes.class_label_ids
    # Access class labels (if available): result.boxes.labels

    # Show the annotated image
    result.show()

    # Save the annotated image
    result.save("output_image.jpg")

🚧 Development Status

⚠️ Repository Under Heavy Development

AngelCV is actively being developed. While core functionality is stable, we're continuously improving and expanding features.

✅ Stable & Ready to Use

• Object Detection: Training, validation, testing, and inference are fully stable
• YOLOv10 Integration: Robust implementation with pre-trained models
• Core API: ObjectDetectionModel and InferenceResult interfaces
• Configuration System: YAML-based model and training configuration
• Model Export: ONNX, TensorRT, and other deployment formats

🔄 Worning On

• Data Augmentation: Expanding augmentation techniques to improve training performance on large datasets
• Performance Optimization: Addressing slightly below-expected performance on big datasets
• Documentation: Comprehensive guides and examples

📋 Coming Soon (TODO)

• Image Segmentation: Semantic and instance segmentation models
• Oriented Bounding Boxes: Support for rotated object detection
• Classification Models: Standalone image classification capabilities
• Additional Architectures: Beyond YOLOv10 (YOLOv9, DETR, etc.)
• Advanced Metrics: Comprehensive evaluation and benchmarking tools

📚 Dive Deeper

For more detailed information, check out our documentation:

• Getting Started: Your first stop for installation and a quick tour.
• Object Detection: Learn about our object detection capabilities, focusing on YOLOv10.
• Configuration: Understand how to use and customize model, training, and dataset configurations.
• API Interfaces: Explore the main Python classes you'll interact with.

🤝 Contributing

Interested in contributing? We welcome contributions of all kinds, from bug fixes to new features. (TODO: Link to contribution guidelines when ready).

🛠️ Development and Support

The primary developer and maintainer of AngelCV is Iu Ayala from Gradient Insight. Gradient Insight partners with businesses to design and build custom AI-powered computer vision systems, turning complex visual data into actionable insights. You can learn more about their work at gradientinsight.com.

📄 License

AngelCV is licensed under the Apache 2.0 License. See the LICENSE file for more details.