Number of stages haar training?
I am doing haar training to identify credential Ids, I have 15 positives and 276 negatives, my processor is a Core i7 with 4 cores/8 threads at 3.4 GHz and 16 gb ram.
I suposed I should use 8 threads and 5.5 gb of ram to each buffer (windows use 5 gb ram), but how about the number of stages, what number will be good to start?
I want that the identification of credentials on images be good, closely to 100%.
"I have 15 positives and 276 negatives" . forget that. try with 1500 positives.
"to identify credential Ids" -- forget that, too. this is the fundamental flaw in your attempt. it only works with a single class of rigid objects
it will never work, the way you want it. it's not about cores, or threads, those classifiers are very restricted to a specific situation (traffic signs, logos, faces, again -- rigid things (of a single kind) !)
please clarify: "identification of credentials" -- what does that mean, exactly ?
The credential Id is the same in all the subjects that I have, is the identification offiicial from mexico, never changes, I supposed that can work, myabe I was wrong, even so I am going to try. Then what are the bases to select a specific number of stages?
again, (might be a language problem) -- what is a "credential" ? (do you have some image / link ?)
"identification of credentials" like card Id, my program must differentiate between a card citizen id and passport
oh, apologies then, mistook it for the "signature"
but again, 1500, not 15 positives, and -- unfortunately, the training process is not optimized at all (single core, no gpu support)
Ok, I only can get 15 beacuse it is extremely dificult get this type of image. Then I must understand that if I tray with 15 the results will be in vain. I think if a set a higer stage number with few positives the result will be better. Then I must forget this problem.