how to create custom cascade for unshapped objects?

edit

Well how to train cascade classifiers is written in the docu - i wont repeat it.

For this kind of problem i recommend training a cnn (convolutionary neuronal network) for object detection (!= classification) There are tons out there you need to understand the concepts a bit and then apply.

Good Luck!

a link to the docs wont hurt.

(but i also think, there's not enough "texture" for a cascade classifier, and the reflections are terrible)

@usuf, can it be, you're running in circles here ?

Well to be honest - training a cascade classifier in today times is just a waste of time imho. Thats another reason why i dont wanted to add a link to the docu. A bit strong opinion -i know.

@usuf, there isn't even a real question here. what are you trying to achieve ? what did you try ?

this also smells quite like an XY-problem.

detecting nuts can probably be done with quite conventional cv methods (shape).

i need to create cascade for my custom object like (toy,my cat,etc) refer some of the tutorial and docs(that u mention above) my doubt is like CNN possible to done object detection using cascade in real time?

(detecting object from real time video not a stable image)

if you have more than 1 class to detect -- for sure forget about cascades, and learn about cnn's.

and sure, cnn detection can be done realtime.

not more than 1 class....for example i need to detect one moving toy from real time is it possible in the cascade ?

you will have to read the paper.

(i won't talk to you, before you proove you did !)

all your ideas about cascades seem to be 3rd hand, and blurry.

no, moving toys cannot be handled like this, neither can your cat.

"I need to create cascade for my custom object" thats exactly an XY Problem. What berak tells you with it - dont make any assumptions about your solution ("cascade classifiers") when asking but describe your use case.

For example: - BAD: "how to train custom classifiers for my detection" - GOOD : "how to detect moving object of type X"

You will get better solutions this way as peopel may suggest things you may never heard about but will solve your problem best. I germany we call this "Look over the edge of a plate".

ON the other hand: If you ask to generic : "How to do x", people could respond with "Not relevant / No real question / Not opencv related, etc". So like always in live, try to find a balance.

F*ck i will read the paper too. @berak Before this dont talk to me anymore, i will proof too - lookin forward!