Modern computer vision requires collecting large-scale datasets, which is expensive . Researchers propose modifications and best practices aimed at minimising human labeling effort . Simulated experiments on a 125k image subset of the ImageNet100 show that it can beannotated to 80% top-1 accuracy with 0.35 annotations per image on average, a2.7x and 6x improvement over prior work and manual annotation, respectively, respectively . Project page: page:

Author(s) : Yuan-Hong Liao, Amlan Kar, Sanja Fidler

Links : PDF - Abstract

Code :

Keywords : image - annotation - cookbook - datasets - practices -

Leave a Reply

Your email address will not be published. Required fields are marked *