Intel presented a new open-source tool for annotating digital images and video. The new tool, called CVAT (Computer Vision Annotation Tool) was developed by researchers at Intel, mainly as an internal asset to help research teams within the company.
Data annotation is a time-consuming process that often becomes a bottleneck in machine learning projects. Building and training large neural network models requires significantly large amounts of data and data annotation represents a real challenge in such cases.
To help overcome this challenge, Intel designed and open-sourced CVAT. The tool supports annotation for supervised machine learning tasks such as object detection, image classification, and segmentation. CVAT provides four types of annotations for images and videos (video frames): boxes, polygons, polylines, and points. This enables labeling data for the purpose of detection and segmentation tasks as well as road marking tasks (using polylines) and face landmarks annotation (using points).
Additionally, CVAT supports integration with TensorFlow Object Detection API for automated annotation. The tool has an accessible browser-based interface, and it provides simple deployment using Docker.
Intel encourages members of the machine learning community, to provide feedback and also take an active part in the future development of the tool. More about CVAT can be read in the official release blog post and the tool is available here.