Hi all. Dr Xin Liu’s group at UIUC has been working on creating a user-interface for our deep learning model, DeepDISC, that is now implemented on the RSP. This effort is lead by our brilliant undergraduate @tiakashyap and also supported by our amazing post-bacc @yaswant2403. The model takes an image and produces an object catalog along with some extra information like object class, photo-z PDF estimates, etc. Our group has created a pretty neat HTML viewer that lets you select predictions and compare our model output to the LSST pipeline detection masks and catalog. It runs in a notebook on the RSP. Since there are no GPUs on the RSP (yet…), the interface is not optimized for large datasets and is mainly designed for examination of individual smaller image cutouts (<=256 pixels). It runs in a few seconds on a single image.
We wanted to advertise this here to folks using the RSP so we created a short video that walks through a demo notebook. Here is the link to the video.
We hope that this tool is useful to a broad user base and would love any feedback for how we can improve or add new features. Please send any comments or questions to Yash at yse2@illinois.edu.