The authors propose a mechanism to characterize object shape from differential motion of the camera with unknown bidirectional reflectance distribution functions (BRDFs). They based their proposal on psychophysical studies, establishing that complex BRDFs do not impede shape perception.
The paper contains a very interesting table summarizing the whole set of results obtained. In general, more restricted BRDFs or lighting provide richer shape information. It is remarkable that at most three motions (images) are required to characterize object shape information.
The paper presents the whole mathematics to get the results. For me, it was hard to follow every case. The case where the method provides an ambiguous reconstruction is described, as it is important to identify cases where the method is not applicable.
Personally, I found it very interesting to apply the results to video online transmission systems and real-time streaming video systems. After reading the paper, I intend on testing the results in real scenarios. For example, the proposal seems to be applicable to provide 3D sense for image capturing in real time.