Initial posting: August 15, 2015
Latest update: Spetember 29, 2015
I no longer need to collect Sony ARW2 sample sequences; the repair tool, KARWY, is available from:
The compression used in the ARW2 raw file formats (at least up to ARW version 2.3), as applied in various Sony cameras, causes loss of precision in some pixel values. That loss is usually negligible, often primarily imposing a texture on noise, but it can become visibly significant depending on the scene and postprocessing. Although the lost precision cannot be recovered, it is possible to make "credible guesses" about what that lost data should be -- and that's the goal of this project. In order to train the repair algorithm to guess better, we need to know what the areas should have looked like for a more perfect sensor: that's the training for which we need your images.
What we need is sets of three Sony ARW 2.3 raw images of a static scene differing only in shutter speed used. Ideally, the images should be aligned at the pixel level, so use a solid tripod or other mount and manual focus. The first should be shot with a "normal" exposure that causes visible artifacts (as you would process the raw). The second shot should be overexposed by enough so that the dark regions have clean detail. The third shot should be underexposed enough so that the brightest regions have some detail and no blooming.
ARW files are large, and upload can take a while, so you will submit the three files one at a time. Uploading a single ARW file from an A7RII takes over 10 minutes using my rather poor home DSL internet connection. If things time out, asking your browser to re-issue the request or using the "back" button may work.
We'll start with the following information and the ARW file that shows the artifacts....
The images you submit here will be used for training the repair tool and by submitting them here, you authorize their use for that purpose. With appropriate attribution, submitted images also might be used in research publications or redistributed as part of a test data set; you will be sent email confirming permission before such use.
You may submit files for as many training cases as you wish by repeatedly using this form.
The repair tool, KARWY, is now available from:
People who have flawed raws still can help by evaluating the quality of repairs performed by KARWY and submitting examples through the WWW form. There is an option in the KARWY WWW form to leave your email address so that you can be contacted for permission to use each photo you upload to the KARWY server.
This is an informal, unfunded, academic research project trying to build a free tool to help the community. Set your expectations accordingly. For more information, contact Prof. Hank Dietz.
The C program that generated this page was written by Hank Dietz using the CGIC library to implement the CGI interface.
The only thing set in stone is our name.