We need a functionable general ObjectRecognition?. We have classifiers based on color and Sift/SURF features. The former is fast but not reliable enough for AtHome?-tasks, the latter is more accurate with multi-color objects featuring textures, but is painfully slow! Furthermore, there is no way of using one and the same application for a general object recognition task. The aim should be to have a modular application that you provide a list of objects along with their corresponding reference-data (i.e. colormap, surf-image/key-file,...) just like the facer does.
Ideally, the application features a 3 stage process:

1. prefilter for region potentially relevant at all with very cheap filter (global colormap?)
2. filter interesting subregions based on individual object-data with cheap filter (color per object)
3. inspect regions with more expensive filter (sift/surf) to determine object class.