AlexeyAB
commented
6 years ago Did you test performance of Yolo v2 detection that built-in latest OpenCV-dnn on Raspberry Pi (ARM)? https://github.com/opencv/opencv/blob/master/samples/dnn/yolo_object_detection.cpp
-
For detection on x86_64 CPU - the maximum performance can be achived using Yolo v2 that built-in OpenCV version >= 3.4: https://github.com/opencv/opencv/blob/master/samples/dnn/yolo_object_detection.cpp More about it: https://github.com/pjreddie/darknet/issues/242
-
For training on CPU - there isn't any plans, because training on the CPU, even with the most powerful optimizations, will take a years
I've observed that NNPack for Darknet is much faster than using the Darknet OpenMP setting on the Raspberry Pi. Do you have plans to integrate this version of Darknet with NNPack for improved multi-threading?