Grosse et al. introduce a dataset for evaluating intrinsic image algorithms consisting of 20 images, each containing a single object on black background. The dataset is available online: cs.toronto.edu/~rgrosse/intrinsic/. In addition, they evaluate several intrinsic image algorithms on this dataset - the python code is also available at the above page. A short script to run explicitly the Retinex-based algorithm can be found at GitHub.
What is your opinion on the summarized work? Or do you know related work that is of interest? Let me know your thoughts in the comments below: