We provide two datasets, one containing lab images, the other containing real-world images. The main difference is that the lab images are exactly in the form as extracted from the camera, while the realworld images have been color-postprocessed with a sharpening software from CVC at Barcelona (more details here) to emulate realistic camera output.

Note that our theoretical model for computing the ground truth holds only for the laboratory images. However, by visual inspection, we found that the ground truth results on the real-world dataset are sufficiently plausible to be used for experiments.

Both datasets and the ground truth are combined in one download link. The resolution of the dataset is 452x302 pixels, to accommodate also computationally expensive algorithms.

The low-resolution datasets can be downloaded from here. (43MB)

Details on the creation of the dataset can be found in the accompanying paper {S. Beigpour, C. Riess}, J. van de Weijer, E. Angelopoulou: "Multi-Illuminant Estimation with Conditional Random Fields", submitted.
If you use the dataset, please cite this paper.