Object detection (other than faces) in OpenCV?
I have been looking for object detection under OpenCV, where I can train for my own set of objects. Only supported feature I found for Object Detection in OpenCV 3.3.1 was WBDetector Class Reference
Looking in further I came across WaldBoost – Learning for Time Constrained Sequential Detection, which in simple words is best or gave best results only on being tested asFace Detection Algorithm
Is there any actual object detection (Other than faces for other objects) algorithm that can be trained and implemented using OpenCV?
what are your objects ? you also seem to have missed this
@berak Simple objects like fruits or vegitables? I only see face detection when I check out object detection in OpenCV nothing more
fruits are difficult ! (they have no reproducable "texture", whatever there is, is usually random)
you'll probably get better results, retraining an existing yolo or ssd dnn network
@berak I want to be sure that there is no other way apart from Neural Networks YOLO and SSD.
@berak I have to convince my professor(who's totally against NN) that Neural Networks are far more worth investing time and resources into.
my argument would be:
none of opencv's builtin detection method looks at color, and they're all not robust against variation in the surface texture
(also, i've yet to meet someone, who used the wbdetector successfully for anything)
@berak Have you ever come across any-other framework other than Neural Network. That has ever detected any object apart from just faces, using image processing concepts?
there are pretrained dpm models in opencv_contrib, for a subset of VOC, like cars,persons,cows, and such, they're quite good ! and the HOGDescriptor comes with 2 prelearned person detectors
@berak What are these Deformable Part Models DPM? Can't seem to find an idea of how these are implemented. I did find a recent paper DeformablePartbasedFullyConvolutionalNetworkforObjectDetection https://arxiv.org/pdf/1707.06175.pdf
https://cs.brown.edu/~pff/latent-rele... for the opencv implementation, look here
(but training still has to be done from matlab, iirc)