Overview

Astro R-CNN is a cutting-edge deep learning framework specifically designed for the analysis of astronomical images. Using the powerful Mask R-CNN model, it excels in detecting, classifying, and deblending sources, making it a vital tool for astronomers and researchers looking to extract meaningful information from complex datasets. With features that greatly enhance performance and usability, this system stands out as a significant advancement in astronomical image analysis.

This tool has seen substantial improvements with the integration of the Detectron2 framework, resulting in a dramatic reduction in training time and increased extensibility. Whether you're working with real DECam images or training your model on personalized datasets, Astro R-CNN promises efficiency and effectiveness in handling a variety of astronomical imaging tasks.

Features

Efficient Source Detection: Utilizes deep learning techniques to quickly identify and classify celestial objects in images, saving significant time in data analysis.
Deblending Capabilities: Excels at separating overlapping sources in crowded fields, which is crucial for accurate astronomical measurements.
Pre-trained Weights: Comes with pre-trained weights for DECam, offering high-performance out-of-the-box results that can be tailored to different data types.
Multi-extension FITS Output: Generates segmentation masks with detailed headers, containing important metadata such as class ID, bounding boxes, and detection confidence scores.
Configurable Training Options: Allows customization of network architecture and training settings, giving users the ability to optimize the model for their specific needs.
Interactive Demonstrations: Provides Jupyter notebooks for users to interactively test and train the model on their own datasets, making it user-friendly and accommodating.
Future Upgrades & Community Support: The project is continuously evolving, with plans for enhancements in user interface, multi-band support, and incorporating community contributions, ensuring ongoing relevance and usability.

Burke86

Astro_rcnn