Benjamin Wu,^1,2* Chao Liu,² Benjamin Eckart,² Jan Kautz²

¹ National Astronomical Observatory of Japan ² NVIDIA

^* Work done as part of NVIDIA AI Residency program

Astronomical interferometry enables a collection of telescopes to achieve angular resolutions comparable to that of a single, much larger telescope. This is achieved by combining simultaneous observations from pairs of telescopes such that the signal is mathematically equivalent to sampling the Fourier domain of the object. However, reconstructing images from such sparse sampling is a challenging and ill-posed problem, with current methods requiring precise tuning of parameters and manual, iterative cleaning by experts. We present a novel deep learning approach in which the representation in the Fourier domain of an astronomical source is learned implicitly using a neural field representation. Data-driven priors can be added through a transformer encoder. Results on synthetically observed galaxies show that transformer-conditioned neural fields can successfully reconstruct astronomical observations even when the number of visibilities is very sparse.

Set up the conda environment using the requirements.txt file:

conda create --name <env> --file requirements.txt

Please replace <env> with any name for the environment you like.

Download the dataset here

Download the model here

Simply run the eval_model.sh script from command line:

sh ./eval_model.sh

To run this, you will need to modify the model, datapath path parameter within the bash script. The script will load the pre-trained model and perform the inference on the test dataset. The results would be saved in the '.../test_res' folder as images.