A deep learning model inference period acceleration method, device and system
A deep learning and model technology, applied in the field of deep neural network learning, can solve problems such as increasing computing time, response delay and device power consumption, and achieve the effect of reducing computing time, reducing device power consumption, and reducing additional overhead
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
example 1
[0089] Example 1: Apply the method of accelerating the inference period of the deep learning model to the smart face capture camera; the smart face capture camera realizes face detection, face key point positioning, and face attributes through the embedded deep learning algorithm Recognition and facial recognition. Among them, the face detection model, key point positioning model, attribute and identity recognition model all adopt convolutional neural network with "batch normalization" structure. The following takes the face detection model as an example to illustrate the process of implementing the method for accelerating the inference period of the deep learning model. The implementation of the present invention on other models can be known by analogy.
[0090] First, prepare the training data of the face detection model; and design and build a convolutional neural network with a "batch normalization" structure and a test data set for the detection task;
[0091] S1: Using ...
example 2
[0099] Example 2: Apply the method of accelerating the inference period of the deep learning model to the cloud analysis server. The cloud analysis server can use GPU, FPGA or other computing accelerators to execute deep learning algorithms for large-scale face identity recognition comparison, pedestrian identity re-identification (ReID), target attribute recognition and video structure in intelligent traffic scenarios. and other functions. Different from smart cameras, convolutional neural networks deployed on cloud servers usually have larger parameters and computing scale. When training large-scale convolutional neural networks, "batch normalization" is essential. Taking the large-scale face recognition model as an example, the implementation process of the present invention is described, and the application of the present invention to other algorithm models can be known by analogy.
[0100] First, prepare the training data of the face recognition model; and design and bui...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More