The world’s largest data set created by Google

In 2016, Google first unveiled a set of data sets for machine learning (Open Images). Google has recently unveiled Open Images Dataset V5.

Because machine learning requires vast amounts of data when learning, Google has released its Open Data V4 (Open Images v4) image data set in 2018. Open Image V4 gives labels and bounding boxes to 9 million images. Open Image v4 is the world’s largest dataset with 1.5 million bounding boxes for 600 categories of objects, with over 300,000 image annotations and annotations annotated. Open Image Data Set v5 is the latest version.

Open Image Data Set v5 features 2.8 million subdivision masks across 350 categories for object instances. Unlike the bounding box, this segmentation mask only recognizes where the entity is. It is a feature that displays the outline of the subdivision mask object and details spatial enlargement. Above all, it contains diverse category objects and illustrations than any data set in the past.

The mask is much more efficient than manual drawing and has an accuracy of 84% in Intersection-over-Union (IoU). In addition to the mask, Google added 6.4 million new human-tested image-level labels, resulting in close to 20,000 categories and 36.5 million labels. It also improves the accuracy of object detection models, including improving tin density in the 600 test and test sets categories. For more information, please click here .