Mathworks introduces Release 2017b (R2017b), which includes the MATLAB and Simulink product families, which greatly enhances MATLAB's support for deep learning and simplifies the way engineers, researchers, and other domain experts design, train, and deploy models. This updated version fully supports the deep learning development process from data annotation, model building, training and inference, and final model deployment. In addition, the biggest highlight of this update of MATLAB is the new component GPU Coder, which automatically converts the deep learning model code to the CUDA code of the NVIDIA GPU. The GPU Coder converted CUDA code can perform inference directly and efficiently from the MATLAB environment. The MATLAB internal benchmarks show that the GPU code generated by GPU Coder is 7 times better than TensorFlow and 4.5 times better than Caffe2.

The media interviewed Chen Jianping, a senior application engineer in China. Chen Jianping started from the data annotation in MATLAB along the development, training and debugging of the deep learning model to finally deploying the high performance model using GPU Coder. It introduced us to MATLAB. This update is for deep learning. Do your hard work. This article will provide a brief introduction to the main points of this update along the MATLAB Deep Learning Development Process, with a focus on the GPU Coder module that automatically converts the model into CUDA code.

Data annotation

For computer vision, the Ground Truth Labeler app in Computer Vision System Toolbox provides an interactive way to semi-automatically label a series of images. In addition to target detection and positioning, the toolbox now supports semantic segmentation, which classifies pixel regions in an image. Chen Jianping said: "Our current annotation tool can directly complete the task semi-automatically. It can automatically mark the pixel level category like the magic wand tool in Photoshop. After we select the image, the tool will automatically pop the object out. After we complete the initialization. After the image semantic segmentation, the tool will use automated means to extract other elements from the subsequent driving process. Because the intermediate and subsequent processes are machine-driven, we only need to use a small amount of manpower in the early stage. Complete the entire labeling process."

This semi-automatic approach does greatly improve the efficiency of labeling, especially for visual system targets such as lane boundaries and car bounding boxes. In this automatic annotation framework, the algorithm can quickly complete the annotation of the entire dataset, and then we only need a small amount of supervision and verification to build an accurate dataset. As shown below, the MATLAB documentation shows us how to create automatic lane markings.

Mathworks R2017b upgrade, enhanced deep learning conversion CUDA code inference speed and then upgrade

We can use different algorithms, such as Auto Lane DetecTIon, which automatically detects lane line features, Aggregate Channel Features/ACF to detect vehicles' ACF Vehicle Detector, and use Kanade-Lucas_Tomasi (KLT) to track one in small intervals. Or multiple ROI algorithms, etc. If we choose an automatic algorithm, then the auto-labeling can be run directly after setting the ROI, maximum lane number, lane line width and other parameters. If the video has been manually fine-tuned and verified, and it achieves good results, we can select "Accept" to complete the labeling task.

AC Contactors

Magnetic AC Contactor is divided into AC contactors (voltage AC) and DC contactors (voltage DC),which are used in power,distribution and power applications.In electrical engineering, because it can quickly cut off the AC and DC main circuits and devices that can be frequently switched on with high current control (up to 800A),Magnetic AC Contactor is often used as a control object for motors, as well as for controlling plant equipment. The electric load of the electric heater, the working machine and various power units, the AC Contactor/DC Contactor can not only turn on and off the circuit, but also has a low voltage release protection. The AC/DC Contactor has a large control capacity and is suitable for frequent operation and remote control. It is one of the important components in the automatic control system.

AC Contactor is our mainly products,we have our own research and development team,we are always commited to provide the good quality,best service to all of our clients.Our AC Contactor had experienced many changes,so you can see that Economic AC Contactor-New AC Contactor-Super AC Contactor,also have LC1-F AC Contactor etc,in order meet different kinds of customer's requirements.

Magnetic AC Contactor,Electrical AC Contactor,Electric AC Contactor,Mini AC Contactor

Ningbo Bond Industrial Electric Co., Ltd. , https://www.bondelectro.com

Posted on