Having Tomasz attached is the only thing that makes me believe that the project isn't total BS (as in they are showing what is easy to give nice results). Tomasz did his thesis on exemplar-SVMs, which is probably the base (or maybe even a more advanced version) of the technology. The biggest turn off is that most CV results are based on many specific conditions, often finding methods turning to the bias of the dataset. Offering a single technology for a wide variety of results is easy to promise (oh it works well on this test set, should work on others), but hard to really deliver. Also many kickstarters won't know the difficulty of PASCAL VOC, and even though PASCAL only asks for a bounding box, they should have made their results more visually appealing.
The software excels with user-supplied specific conditions; which might be a feature.
The technology is pretty cool but the concept of kickstartering this is fairly odd to me. Why not try and raise equity/debt capital around the concept of computer vision as a middleware/service?
Well, kickstarting does a couple things: it validates the business model and (potentially) gets a seed round without giving out equity or taking on debt.