I am using a BOW classifier for classifying my input images (motocycles (front, back, lateral, details), cars (front, back, lateral, details), interiors (front, back, lateral, details), or other stupid things (commercials, trees, rooms, etc)). I have a question how big shall be the datasets used for training? I am asking this, because it seems not to evolve very much on each increase of 100 images per class. Right now I have 400 in each class. Any suggestions?
This forum is disabled, please visit https://forum.opencv.org