Google on Monday announced the open source release of the company’s latest AI-powered image segmentation model – DeepLab-v3+. Essentially, this model has been put in place to create a “synthetic shallow depth-of-field effect” like the one shipped with the Pixel 2 and Pixel 2 XL. Opening up the software for developers, Google aims at enabling third-party apps to create software-based depth effect images. Additionally, the DeepLab-v3+ model will be built on top of convolutional neural network architecture and implemented in TensorFlow.
According to the blog post by Google Research, the tech giant is opening up the availability of its proprietary model to allow app developers to create bokeh-style image effects where object outlines are pinpointed for sharper image blurs. Google Research Software Engineers, Liang-Chieh Chen and Yukun Zhu, state that this technology has “much stricter localisation accuracy requirements than other visual entity recognition tasks such as image-level classification or bounding box-level detection.”
Most other OEMs including Apple and Samsung make use of dual image sensors to create the same effect that Google has been offering with just a single camera on the Pixel 2 and Pixel 2 XL. The implementation is so accurate that the Pixel 2 has emerged as one of the top rankers in DxOMark’s camera ratings.
“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,” says the blog post.
Do note that the company has issued an update to say that this is not the exact technology currently available on the Pixel 2 range of smartphones. We can, although, expect it to produce similar results.
[“Source-gadgets.ndtv”]
