Open mzillich opened 7 years ago
Can you clarify that please? How and where does recognition depends on segmentation?
The do_recognition() is only run on segmented objects. This was fine for the last demo to speed things up. But is not always what one needs.
We have to rethink the structure of squirrel_look_for_objects.cpp. Right now it segments the scene first and calls the recognizer on segmented objects. It would be better to have the recognizer running on the scene and then mask out the recognized objects and do a segmentation afterwards (as the recognizer is more robust).
Recognition should not depend on segmentation, at least not always.