IBM has released a new free tool designed to help developers cut the chore of labeling images in video when training AI object-detection models.
The new auto-labeling tool is part of IBM's push to stay relevant to developers building AI products using the Google-built open-source TensorFlow framework in the IBM Cloud.
IBM hopes its auto-labeling tool attracts developers looking to save effort and time when developing new AI applications with TensorFlow. IBM estimates it takes 200 to 500 samples of hand-labeled images for an AI object-recognition model to detect a specific object.
The promise is that developers building these models can spend time on more interesting and valuable tasks than labeling objects in video, which is as tedious and as combing through receipts at the end of the month – but also necessary.
The tool is part of IBM's open-source Cloud Annotations image toolset that offers developers a fast track to building object-detection models.
"Autolabeling images speeds the process and gives developers back valuable time to work on other innovative projects," said Nicholas Bourdakos, a developer advocate at IBM's Cognitive Applications.
Other tech giants are also tackling the labeling challenge. Microsoft has released an open-source tool for labeling objects in video called the Visual Object Tagging Tool. AWS launched Amazon Rekognition Custom Labels in December, aimed at helping businesses spot objects and images that are relevant to their business.
The IBM auto-labeling feature is in an "early beta", according to Bourdakos, but developers can use it to model a project with automated labeling.
The feature is available on GitHub now by uploading video footage to IBM's Cloud Annotations tool.
Bourdakos recently posted a video demonstrating how the auto-labeling feature works on video footage of two dudes showing thumbs-up and thumbs-down.