2011 Aug 17: We are beta on the web!
As we hoped, our awesome GSOC interns Chen and Lalimarmo fixed up our web presence, added openID authentication, and created a data-sharing and idea-sharing system for our users. We are now running open on the web; if you want to check it out, see our use-the-code page.
2011 Jun 06: The start of our Summer of Code.
Google Summer of Code interns Kevin Chen and Carlos Lalimarmo started work on's web services today, with a meet-up in New York City. Chen and Lalimarmo made a pretty hard cut—we had many GSOC applications from around the world—so they are very capable open-source developers. The goal is to move from alpha to beta and have a proper API, along with much more (and much more fun) functionality for our users. Go team!
2011 Apr 01: Comet 17P/Holmes.
We released a somewhat light-hearted paper and got some good press, notably Wired and New Scientist.
2011 Mar 24: We're doing Google Summer of Code.
Interested in applying? Contact gsoc at for details.
2010 Sep 02: Feige 85 re-found.
Rare white dwarf Feige 85 was published in 1958 with an approximate and slightly worng position, causing it to be lost to white dwarf researchers forever. Today Doug Welch (McMaster) and friends put a grabbed image of a finding chart into and re-found it! It turns out the star is (boringly named and unrecognized) 2MASS J13362125+0822335. See Slacker Astronomy for more details. Later they also recovered lost M7 star HaroChavira 62 by the same method.
2010 Jan 12: Sam Roweis, 1972–2010.
Sam Roweis (NYU, Google, Toronto), the co-PI of, died today. He had an immense impact on machine learning and computational statistics, but also on astronomy, where his contributions—including—were truly disruptive. He was a great scholar and a great friend.
2009 Oct 14: paper finished.
The long-awaited paper on finally appears on astro-ph as arXiv:0910.2233. We are also submitting it to AJ. Please cite this paper (or the AJ version, when it appears) in any work that uses Thanks!
2009 Sep 11: Lang successfully defends his PhD.
Dustin Lang's thesis is on At the defense are Sam Roweis, David Hogg, Sven Dickinson, Rich Zemel, Doug Finkbeiner, and exam chair and, coincidentally, UofT astronomer Bob Abraham. A good time, and a bit of champagne, was had by all.
2009 June 26: DeepSky processes 14 million images; PTF adopts
Peter Nugent of the DeepSky team wrote us email summarizing his results with an enormous data set: For DeepSky we have successfully processed 14M images, and we just finished up with all of this the past week! This was ∼70TB of data we went through in the past year. The images were 2048X600 pixels with 0.858 arcsec/pix... This is 5 years of all-sky data (∼20,000 sq deg/yr) hit quite a few times. In the same email he went on to say that the Palomar Transient Factory has adopted in its pipeline processing.
2009 February 18: Interview with Flickr developer blog
In a blog post on the Flickr developer blog Christopher was interviewed about topics ranging from how amateur astronomers are contributing to science, the workings of, and the experience of hooking it all up to Flickr. This led also to our first appearance on Slashdot.
2008 November 6: Flickr group makes front page of reddit
In addition to being featured in reddit, we were also covered in kottke, and O'Reilly's Make magazine blog.
2008 October 21: Hogg and Dustin propose theory of everything
Automated calibration (that is, what the service does) is the first step towards automated joint scientific analysis of heterogeneous imaging: automated science, as it were. We wrote this up for a European workshop on data analysis. Check it out at our bibliography.
2008 July 6: Flickr group adds annotations
Images solved in the flickr group are now annotated. Major objects are being labeled on new images being submitted to the group. An example on a picture of M42.
2008 May 7: Blind Date paper finished.
The paper documenting our new system for determining the date-of-origin of historical imagery using only pixel data and the known proper motions of catalog stars has been finished. Check it out on the arXiv or at our bibliography.
2008 April 4: We added meta-data to a video.
The FIREBall balloon-borne UV telescope put a video from its guider camera on YouTube. We ripped the video to images, annotated the images with meta-data, and re-built the video. Watch it here.
2008 March 20: Clean USNO-B data released.
The data generated by our "Clean USNO-B" project has been released. Find it here.
2007 September 12: Version 0.1 of the code is released.
All the code used in the alpha-test web service, all the code used to solve the Astronomy Pictures of the Day, and all the code used to solve and recover SDSS and GALEX data has been released to code requestors. If you would like to obtain the code for yourself—or become an alpha-tester—please hit the use menu link.
2007 September 10: We recovered lost SDSS data on Messier 71.
Substantial telescope data acquisition system glitches caused the Sloan Digital Sky Survey automatic astrometric calibration pipelines to fail on the imaging run covering Milky Way globular cluster Messier 71 (among many other things). astrometrically calibrated one field from the SDSS run blind (that is, with no prior information about pointing), and the SDSS pipelines were then able to bootstrap solutions for the entire run off of that one solution.
2007 July 29: Flickr group created.
An astrometry group was created on Flickr. Images submitted to the group are solved in a fully automated fashion using the blind solver and the Flickr API. Any Flickr user can submit to the group.
2007 April 16: We are live! Alpha testing has started.
We have started alpha testing with a limited number of development collaborators who will help us work out the kinks and improve the system. We are reliably solving images from about 10 arcmin up to 100 degree scales including over a hundred examples from the Astronomy Picture of the Day. (We are still not producing high precision WCS and are still solving all images "blind", but both of those will change soon.) Our next milestone will be to put out a code release which will allow users to compile (and modify) the entire system on their local hardware.
2007 February 14: We solved today's Astronomy Picture of the Day blind.
In preparation for going alpha, we have been running random images through the system. We added jpeg capabilities so we can deal with web images as well as professional images. To test the system, we ran the Astronomy Picture of the Day for Valentine's Day, which was, of course, of the rosette nebula. We solved it—that is, automatically determined its location on the sky and orientation—right out of the box.
2007 January 18: We solved 7077 GALEX/NUV fields blind.
We ran 7077 fields from the GALEX All-Sky Imaging Survey, near-ultraviolet (NUV) channel, through a cascade of blind solving attempts and got all but 8 of them, with no false positives, for a success rate of nearly 99.9 percent. We built our indexes with a blue-biased subsample of the USNO-B1.0 catalog. The GALEX far-ultraviolet (FUV) channel is much more challenging because only a small fraction of FUV sources can be associated with sources in the USNO-B1.0 catalog.
2006 December 4: We solved 35,000 SDSS u-band and z-band fields blind.
The blind solver—which finds WCS for images with no meta-data—is optimized for r-band images. However, we discovered today that it works pretty well in other bands. Without any changes to account for band differences, our system, which now gets 99.6 percent of SDSS r-band fields right (with no false positives), gets 96.9 percent of u-band fields (with no false positives), and 99.3 percent of z-band fields (with no false positives). This gives us some confidence that a multi-wavelength system will be possible, though with the exception of some GALEX testing, we have not yet ventured outside the optical.
2006 September 28: Roweis presented the current status of the project at Google Pittsburgh.
Roweis's presentation focused on the computer-science and machine-learning aspects of the project. You can read the presentation materials here (ppt) or here (pdf).
2006 July 2: We solved 336,554 SDSS r-band fields blind.
Today we ran a set of 336,554 SDSS fields through the blind solver, and have a total success rate higher than 99 percent. We don't know the exact success rate yet, because many of the 1,976 failures are fields that are out-of-focus, inside small holes in the USNO-B1.0 catalog, or subject to other problems for which we are not responsible. (Yes, we are solving a subset of all of the fields, not just the science-grade fields!) The entire set of fields solves overnight on a University of Toronto computer cluster.
2005 October 26: We solved our first GALEX/NUV field blind.
Today we re-solved the astrometry (ie, we determined the pointing, rotation, and scale) for a single GALEX near-UV image from the All-sky Imaging Survey using the x,y positions of sources from the GALEX pipeline and the USNO-B1.0 catalog and nothing else. The source positions were given to us by David Schiminovich (Columbia) who withheld all information about the image pointing, rotation, and scale. The only thing we assumed is that the image is larger than about 30 arcmin in diameter.
2005 September 5: We achieved robust WCS optimization (prototype).
We have created a system (not yet even ready for beta release) that takes an x,y list of compact sources in an image and an approximate WCS and produces as precise WCS as is possible given the USNO-B1.0 catalog. The WCS is fit including Spitzer Imaging Polynomial distortions to the user-chosen SIP polynomial order. This code works on SDSS images (fields), typical HST/ACS images, and (we have learned since) GALEX images.
2005 July 28: We solved our first SDSS r-band field blind.
Today we re-solved the astrometry (ie, we determined the pointing, rotation, and scale) for our first SDSS r-band image (field) using the x,y positions of stars from the SDSS pipeline and the USNO-B1.0 catalog and nothing else. We took the stellar positions from the SDSS pre-calibration pipeline, so we had no access to the RA, Dec positions nor even the calibrated magnitudes of the stars we used. The only things we assumed were (a) that the field is in the North Galactic Cap (one-quarter of the sky) and (b) that the image is of order 10 arcmin in diameter.