This benchmark is done in python using google benchmark.

We are benchmarking results from two different models (yolov5 and FasterRCNN), six different images (dog.jpg, eagle.jpg , giraffe.jpg, horses.jpg , kite.jpg, person.jpg ) and six different algorithms (nms-simd, tensorflow, torchvision, faster-nms, opencv, mmcv). Current results are taken from a Ryzen 2600 machine.

google-benchmark onnx-runtime tensorflow torchvision opencv mmcv numpy

Weights should be in root project folder. Download FasterRCNN. Clone yolov5 whereever you want then export as onnx. Default weights for yolov5 is yolov5x6.

First run one of the object detection scripts (yolov5.py or fasterrcnn.py). Then run benchmark.py. Then run the other object detection script, rinse and repeat.