Effective multiple object detection with a small data set [closed]
Hello,
I want to detect objects according to images from a database filled by the user, with a maximum amount of 6 images(for 6 different angles).
This amount is not enough for building a classifier, but I still need to be able to detect the objects in different backgrounds, different light conditions etc, in real time.
I also need to detect multiple objects in every frame and be able to find a relation between the detected objects (for example: Cup on Table).
How can I perform this kind of detection when I only have a small data set that I can't control? How sould I apply it on the other objects without increasing the response time? And How could I recognize the relation between them?
Thank you.
good luck ;)
It all depends on what accuracy you want to reach but in my opinion, challenging and probably impossible. Try to define beforehand if the object is prone to have deviations. Because if it is a known cup, than it is a recognition problem and not a detection problem and your chances are slightly increasing! Keep in mind that results on the theme of single instance object recognition are still quite low. A bag of visual words approach might do the trick ...
Hi, thanks for your quick response.
In my dattabase I keep the name of the object with it's images, so connecting between an image and it's name is an easy task. I thought that detection might be easier to perform, and thats why I wrote detection instide of recognition. As for the accuracy, I need it to be as high as possible on any object in the kitchen. Do you think Bag of Words is the best method to use here? Thanks again!
"Do you think Bag of Words is the best method to use here ?" - this will be helpful for sure, as long as you can "pre-train" the dictionary on a huge dataset.
Yeah thats the downside of BoW ... the pretraining of the dictionary.