Inverted files not used in Vocabulary Tree find ?
Hi,
I found the implementation of the vocabulary tree in ROS very interresting, but by looking carefully at the code I've noticed that the inverted files are not used to query the document database.
As far as I understand it, this kills all the interest for a visual vocabulary tree. Indeed, if i have to scan ALL THE DOCUMENTS in the database to find the matches (as it is actually done), why bother quantitizing features in words ? I could as well compare directly the query image with all the images in the database.
But the code is nice. If no further development is planed to make a proper use of the tf-idf files that allow a FAST COMPUTATION of the distance between the query document and the database, I'm willing to contribute.
Patrick, would you like us to work together on it ?