Skip to main content

How can I find morphologically similar images in the Science Archive? - Knowledgebase / Archive & Data Retrieval - ALMA Science

How can I find morphologically similar images in the Science Archive?

Authors list

It is possible to find morphologically similar images in the ALMA Science Archive by looking for the

tildeicon.png

icon next to the preview images on the Archive Query interface or on the Request Handler. Clicking on the icon will bring up an interface showing the 1000 most similar images to your selected image. This works for continuum images as well as for data-cubes where we make use of the peak-flux (moment8 ) images. On this interface, you can now select additional images of interest by clicking with the mouse (or right-click to select all images up to a given one). Each time you do so, the interface instantaneously updates the display, reordering the remaining images by their simultaneous similarity to all images selected so far. The automatic reordering can be toggled off and on using the control on the right-hand side of the top bar. Hovering over one of the thumbnail images provides additional information about the corresponding observation and FITS file. There are five ‘quick-look’ images displayed at the top right which correspond to the five largest image groups in the list of 1,000 images. Click on one of them to pre-select similar images in the interface.

After making a selection, you can send all the FITS files corresponding to selected images to CARTA with a single click (follow the instructions at the top of the CARTA pop-up to see all images there and maybe use the top right icon to open the display in a new browser tab), to get a download script for those FITS files, or to open a new ASA query tab with the observations corresponding to the selected images. In the new query tab you can further filter the selected observations using all criteria available there.

A description of the state-of-the-art deep learning method used to determine similar images - self-supervised contrastive affine-transformation-independent representation learning with a deep neural network - as well as a more in-depth description of the interface can be found in this ESO Messenger article.

 

 

asa_image_similarity_news_item.png