It is a tricky question : How choose the visual words in order to gather the key points in an image ? Do I do a clustering on all the observed key points in the database ?
This solution sounds good, however, it needs to have all the key points in the memory at the same moment... I will try to have different pointers on each key points and cluster the key points on each image (1000 by image) before clustering them in 500 sub clusters for all the database. I have no idea if it is enough, testing with large dataset will say us !
If you have other idea, it would be interesting to speak about it
It is a tricky question : How choose the visual words in order to gather the key points in an image ? Do I do a clustering on all the observed key points in the database ? This solution sounds good, however, it needs to have all the key points in the memory at the same moment... I will try to have different pointers on each key points and cluster the key points on each image (1000 by image) before clustering them in 500 sub clusters for all the database. I have no idea if it is enough, testing with large dataset will say us ! If you have other idea, it would be interesting to speak about it