Semantic image segmentation is the task of categorizing every pixel in an image and assigning it a semantic label, such as “road”, “sky”, “person” or “dog”. And now, Google has released its latest image segmentation model as open source, making it available to any developers whose apps could benefit from the technology.
The function can be used in many ways. One recent application in the world of smartphones is the portrait mode on Google’s latest Pixel 2 devices. Here, semantic image segmentation is used to help separate objects in the foreground from the image background. However, you could also imagine applications for optimizing auto exposure or color settings.
This kind of pixel-precise labeling requires a higher localization accuracy than other object recognition technologies, but can also deliver higher-quality results. The good news is that Google has now released its latest image segmentation model, DeepLab-v3+, as open source, making it available to any developers who might want to bake it into their own applications.
Modern semantic image segmentation systems built on top of convolutional neural networks (CNNs) have reached accuracy levels that were hard to imagine even five years ago, thanks to advances in methods, hardware, and datasets. We hope that publicly sharing our system with the community will make it easier for other groups in academia and industry to reproduce and further improve upon state-of-art systems, train models on new datasets, and envision new applications for this technology.
If you are interested in finding out more about DeepLab-v3+, head over to the Google Research Blog for more details.