Google has taken a first exploratory step towards making image annotation faster and easier. It has developed a new a machine learning–powered interface, Fluid Annotation, for annotating the class label, the outline of every object and the background region in an image. The solution is said to accelerate the creation of labeled datasets by a factor of 3x.
Traditional manual labeling tools require an annotator to click on the boundaries carefully to outline each object in the image. This is a tedious process. Human annotators would require 19 minutes to label a single image in the COCO+Stuff dataset, and 53,000 hours to label the whole dataset.
Fluid Annotation starts from the output of a strong semantic segmentation model, which a human annotator can modify through machine-assisted edit operations using a natural user interface. Google’s interface empowers annotators to choose what to correct and in which order, allowing them to focus their efforts on what the machine does not already know.
Click here to continue reading on how the new solution from Google labels pictures and helps human annotators enrich the labels.
Please give your feedback on this article or share a similar story for publishing by clicking here.
