The Complete Magazine on Open Source

Baidu develops open source deep learning framework for smartphones

1.26K 0


Chinese Internet giant Baidu aims to get bigger in the world of artificial intelligence (AI) space by launching its open source mobile deep learning framework. Called Mobile Deep Learning (MDL), the new development is a convolution-based neural network that uses the image sensor on Android and iOS running mobile devices to recognise objects.

Baidu’s MDL has planned the neural network with low complexity and high speed. These are the certain parameters to enable deep learning on mobile devices. But interestingly, the new model is also designed to use GPU on iOS devices to enable calculations through the network. This makes Apple devices a step ahead of Android models.

A couple of demonstrations provided by Baidu on the GitHub repository shows its different areas. The first demonstration reveals that MDL is capable of recognising an object in less than a second’s time. The second demo, on the other hand, shows the project’s ability to identify an object from an image and find its suitable matches from the Web.

While MDL does support developments produced on UC Berkeley’s deep learning framework Caffe, Baidu recommends developers to use its PaddlePaddle platform that was launched last year. You can convert your existing PaddlePaddle model to an MDL project using simple Python skills. Furthermore, Baidu has given support for projects such as MobileNet, SqueezeNet and GoogLeNet.

MDL is available on GitHub under the MIT License.