Closed zanazakaryaie closed 5 years ago
Simd does not support OpenCV HOG models. Although Simd contains primitives for optimizations of HOG detector. But I have opinion that using of HOG/HAAR/LBP detectors is not best way to achiev of best accuracy and performance now. In present there is only one good way for object detection - neural networks.
Sure. DNNs are generally better than classical algorithms that work based on hand-crafted features. But it all depends on the processor power. I'm using a low-end ARM processor. Even the fastest deep object detectors like Yolo or SSD are too slow for my application. I'm looking for a lower accuracy but much higher performance.
Lately I develop a project to fast inference of previously trained neural models. It is something like Inference Engine but it also supports ARM CPU. May be it will be useful for you. For our purposes (faces and licence plates detection) on mobile CPU its performance is good enough.
Hi again. Thank you very much. I didn't know that. I will test it as soon as possible. Thanks again :)
Hi
I want to train a HOG+SVM detector in OpenCV and then use your library for acceleration. Does it support both HOG and SVM?