Shulou(Shulou.com)11/24 Report--

According to CTOnews.com on January 18, Intel's open source computer vision library OpenCV 4.7 was officially released in December last year. The new version brings a new ONNX layer, greatly improves the convolution performance of DNN code, and provides Huawei Teng CANN backend support for DNN modules.

According to the latest test of OpenCV China team, based on the acceleration ability of CANN, the CANN back-end is tested with PP-ResNet50, MobileNet and YOLOX in OpenCV Zoo. It is found that the three models achieve the excellent results of 3.29ms 1.21ms-12.80ms respectively under the CANN back-end.

At the same time, the reasoning result of the CANN backend is basically consistent with that of the default CPU backend. With OpenCV, users only need the following 7 lines of Python code, and without learning the API of CANN, you can simply call the Penton AI processor to realize neural network accelerated reasoning (the core is the third and fourth lines of code, and the rest are common model reasoning logic codes):

Import cv2 as cvnet = cv.dnn.readNet ("/ path/to/model.onnx") net.setPreferableBackend (cv.dnn.DNN_BACKEND_CANN) net.setPreferableTarget (cv.dnn.DNN_TARGET_NPU) input = cv.imread ("/ path/to/image.jpg") net.setInput (cv.dnn.blobFromImage (input)) out = net.forward () ▲ example code using CANN back-end reasoning teacher Yu Shiqi of the computer Department of Southern University of Science and Technology said Even if OpenCV has winograd conv and other optimizations for ARM CPU, it takes more than 20ms to complete ResNet50 reasoning on Apple Benchmarking CPU M1, while Huawei Teng only uses 3.29ms.

CTOnews.com learned that Teng CANN (Compute Architecture for Neural Networks) is a heterogeneous computing architecture launched by Huawei for AI scenarios. Teng CANN supports neural network reasoning and training frameworks such as MindSpore,OpenCV DNN, Paddle, PyTorch, TensorFlow, etc., and is compatible with a variety of underlying hardware devices.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.