This archive contains all the captured RAW/JPEG pairs (all 140 patches under all 16 illuminants across different exposures) for each of our calibrated cameras. The data is stored in MATLAB .mat files with the following variables:
|R,J||3x140xN matrices containing corresponding RAW and JPEG values (in the range [0,1] and integers in [0,255] respectively). The first dimension indexes color channel (sensor space for RAW, and RGB for JPEG), and the second indexes the different patches in the color chart. The final dimension indexes the different images of the chart captured for each camera, each under a distinct exposure and illuminant pair.|
|iTag||Nx1 matrix containing the ID of the illuminant for each image, i.e., all the values R(:,:,i), J(:,:,i) correspond to RAW, JPEG colors of the chart captured under illuminant iTag(i). This ID takes integer values in the range [1,16], where the illuminants are optimally ordered to maximize the convex hull of the captured chromaticities as described in the paper. Therefore, to measure calibration quality with say only four illuminants, you should only consider measurements for which iTag takes values 1,2,3, and 4.|
|eTag,exp||Nx1 matrices containing the exposure ID and (relative) exposure time of each image. There are multiple images captured under each illuminant with different exposures.|
|nfact||Scalar value included only for the JPEG only cameras. It corresponds to the normalization factor applied to the RAW proxy's sensor measurements to generate R, and is a parameter that needs to be passed to the calibration function (see the code documentation for details).|
Use the selSubset function included with the code distribution to select different subsets of this data. For example, the following code selects returns all of the captured data in the matrices rall,jall:
>> c = load('/path/to/datafile.mat'); >> [rall,jall] = selSubset(c,0,[1:max(c.iTag)]);
Note that given these files, you will still need to call InvertBase and InvertBaseQ to generate the precomputed files required for computing the probabilistic inverse with invJ.