Using deep learning for Detection, Instance Segmentation, and Classification on astronomical survey images.
Reference Paper: Merz et al. 2023
Corresponding Author: Grant Merz, University of Illinois at Urbana-Champaign
Contributors (in alphabetical order): Patrick D. Aleo, Colin J. Burke, Yichen Liu, Xin Liu, Grant Merz, Anshul Shah, .
This is an updated repo of the original implementation (https://github.com/burke86/astro_rcnn)
DeepDISC is a deep learning framework for efficiently performing source detection, classification, and segmnetation (deblending) on astronomical images. We have built the code using detectron2 https://detectron2.readthedocs.io/en/latest/ for a modular design an access to state-of-the-art models.
Setup:
conda env create -f environment.yml
You will also need to install scarlet and detectron2. Building from the source is recommended for both
Usage:
demo_decam.ipynb
This notebook demonstrates how to set up, train and evaluate a model using the detectron2 API. It requires the user to have downloaded the PhoSim simulated DECam data used in Burke et al. 2019, MNRAS, 490 3952.. The data can be found here: training set (1,000 images) validation set (250 images) test set (50 images).
demo_hsc.ipynb
This notebook follows a very similar procedure to demo_decam.ipynb, but for real HSC data. The ground truth object locations and masks are constructed following training_data.ipynb and classes are constructed with external catalog matching following hsc_class_assign.ipynb It is largely for demo purposes, so will not reflect the results of the paper. The training scripts we used to recreate the paper results are in train_decam.py and train_hsc_primary.py