What is SSD in object recognition

Machine learning: Google publishes an API for object recognition

Recognizing objects in images is an application of artificial intelligence that is equally practical and well suited for demonstration. With the TensorFlow Object Detection API, Google has now released a system that the company has been using internally for some time, according to the research blog. Among other things, it is available as part of the tool for the Nest-Cam devices, which can, for example, provide a notification when a person steps into the camera area. The search engine provider also uses it to recognize street names in Street View and to find style suggestions in the Google Android app.

The open source framework is based on the TensorFlow ML (machine learning) framework promoted by Google and already contains several detection models, including a single shot MultiBox Detector (SSD), region-based convolutional neural networks (R-CNN) and region -based Fully Convolutional Networks (R-FCN) as well as extensions based on them. A Jupyter notebook is also part of the open source project. The MobileNets variant of the SSD is particularly slim and therefore optimized for mobile use. Google won Microsoft's COCO (Common Objects in Context) Detection Challenge with a version of the Faster R-CNN 2016.

Image analysis at Google and its competitors

In September of last year, Google had already released "Show and Tell", a system for creating captions that recognizes and describes objects and is also implemented as a TensorFlow model. In the summer of 2016, Facebook published DeepMask and SharpMask open source libraries that recognize objects within images and are based on the TensorFlow alternative Torch. Microsoft also offers image analysis functions with Cognitive Services, as does IBM with the Watson API. (rme)

Read comments (47) Go to homepage
Ad ad