The AI ​​behind Pixel's super-high portrait mode is launched by Google into open source

Camera on Google's Pixel phone is one of the best cameras available today, and one of the main reasons is the AI ​​system behind it. Google uses machine learning technology to select the best pictures. And now, they are planning to turn one of these AI tools into open source tools.

Posted on the blog earlier this week, Google launched the open source version of DeepLab-v3 +. This is an image partitioning tool using Convolutional Neural Network (CNN). A machine learning method is very good at analyzing image data. Partition the image to analyze the object in the image, divided into regions, split the image in the foreground and background in the image.

How the image partition works

While it may sound normal, this method is very useful for cameras and Google uses it for its Pixel portrait mode. The result was bokeh-style images, blurring the background but the subject was still clear.

The iPhone itself made the technology popular, but it should be noted that Apple uses two lenses to create a portrait effect, while Google uses only one. (And the portrait mode of Apple or Google is better, readers should judge for themselves).

According to the explanation of software engineer Google Liang-Chieh CHen and Yukun Zhu, image partitioning has improved rapidly thanks to the flourishing of deep learning, reaching 'unimaginable accuracy in the past 5 years'. . Google hopes to share widely with 'other groups in the industry to continue to improve' the work they have done.

Bringing such software to the community will help application developers when they need image partitioning technology, and catch up with Google.

https://github.com/tensorflow/models/tree/master/research/deeplab

See more:

1. This is a portrait shot that removes fonts (Portrait Mode) on Pixel 2 without Pixel 2
2. This is why you should buy Pixel 2 instead of iPhone X
3. Compare the best photography smartphones in 2017