This page provides a new rendering of the Colour Chart dataset provided by Peter V. Gehler and other authors. The dataset contains 568 images taken around Cambridge, England on two DSLR cameras (Canon 1D & Canon 5D). The images were captured in Canon RAW format. The authors provide TIFF renderings, output by the Canon Software. These renderings are visually pleasing, but they are not linear. You can find out more about the dataset here:
http://people.kyb.tuebingen.mpg.de/pgehler/colour/
Since a linear rendering is important for a lot of Colour Constancy algorithms Lilong Shi and Brian Funt provided a linear rendering of this dataset which you can find here:
http://www.cs.sfu.ca/~colour/data/shi_gehler/
Why Rerender Again?
We noticed that the renderings provided by Shi and Funt made the colours look washed out. There also seemed to be a strong Cyan tint to all of the images. Therefore, we processed the RAW files ourselves using DCRAW. We followed the same methodology as Shi and Funt. The only difference is we did allow DCRAW to apply a D65 Colour Correction matrix to all of the images. This evens out the sensor responses. This transform is linear, so would have no negative affects for image processing. Three of the images are displayed side-by-side below:
The images have been raised to the power of 2.2 to correct the Gamma for display. On the left is the Shi and Funt rendering and on the right is our new rendering. We examined the transform between the two images, and it appears to match the black level correction that DCRAW does. Consequently, the authors did not correct for the black level as they had claimed. The rendering they provided is linear, but the pixels are a translation away from the actual responses. This causes the "washed-out" effect. The translation means that varying shades of colours will also vary slightly in chromaticity. This is a problem when using the grey-patches on the Macbeth Colour Chart as a gage for the scene illuminant.
The Rendering
This version was rendered in document RAW format, which had no demosaicing. The images were output as linear TIFFs. Here is the DCRAW command we used:
dcraw -d -v -4 -o 0 -T -w IMG_0001.CR2
This output greyscale images, because the pixel responses mapped directly to the bayer-pattern on the camera's sensor. We demosiaced the images manually in Matlab. This was done by averaging the two Green responses at each 4x4 pixel, then taking the unchanged Red and Blue responces to make an RGB. This resulted in an image half the size.
We provide only the 482 images taken on the Canon 5D camera. We recomend using only these images for consistency, since they were all taken on the same camera. This means that the error measure used accross images can be fairly compared.
Download
Linear 16bit TIFFs (split into six zip files, each zip file is about 1GB):
First | Second | Third | Fourth | Fifth | Sixth
Linear 16bit TIFFs with the Macbeth Colour Charts Cropped out (split into six zip files, each zip file is about 1GB);
First | Second | Third | Fourth | Fifth | Sixth
Linear 16bit TIFFs, Thumbails with Macbeth Colour Charts Cropped out. Smaller images at 549 x 365 pixels. Each image is about 1mb.
Additional data. This zip file includes a matrix of the Macbeth Colour Patch RGBs for each image as a .txt and as a Matlab .mat file. The 24 Macbeth Patches were cropped out of each image and the average RGB was taken for each one. It also includes the illuminant RGBs as a .txt and a .mat file. The illuminant RGBs are calcuated as the average of the non-clipped grey patches on the Macbeth Colour Chart.
Referencing
If you would like to use this data please provide the following reference:
Stuart E Lynch and Graham D Finlayson, UEA Colour Group, "Reprocessed Linear Version of the Gehler dataset, with Black Level Corrected" accessed from http://colour.cmp.uea.ac.uk/datasets/.
Also, please also cite the original work:
Peter Gehler and Carsten Rother and Andrew Blake and Tom Minka and Toby Sharp, "Bayesian Color Constancy Revisited," Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2008. and http://www.kyb.mpg.de/bs/people/pgehler/colour/index.html.