Exploring how weak supervision can assist the annotation of computer vision datasets

Abstract

Current artificial intelligence (AI) workflows depend on researchers performing laborious annotation work. In the case of computer vision, crowdsourcing is a popular alternative to alleviate this effort. The general public can provide even more trustworthy annotations with the help of existing frameworks. This paper proposes an image dataset annotator helper that uses weak supervision and explains how class activation maps (CAMs) are integrated with deep image classifiers to produce weakly supervised localisers that could further improve human image annotation performance. Comparing these models with primary crowdsourcing data revealed that the models can annotate better than humans by 9.7% when measuring the localisation error while taking into account both false positives (FPs) and false negatives (FNs). Moreover, the models can also save up to 36% of the time required to perform manual image annotation. This confirms that there is potential within CAM-empowered models to further improve the image annotation experience.