How Deep Learning And Semi-Automation Improves Image Annotation
By INESC Staff: Luis Vilaça, Paula Viana, Pedro Carvalho and Teresa Andrade
A key to FotoInMotion becoming a powerful tool is in the efficiencies gained by automation, which will be achieved in part thanks to cutting-edge developments of INESC, a leading research institute based in Porto, Portugal.
INESC presented their paper “Improving audiovisual content annotation through a semi-automated process based on Deep Learning” at the 10th International Conference on Soft Computing and Pattern Recognition. The conference was an opportunity for researchers to discuss the latest solutions and scientific results and methods in solving intriguing problems in the fields of soft Computing and Pattern Recognition.
The main objective of the work presented by INESC at SoCPaR, which was held in Porto, December 13-15, is to semi-automate the process of image and video content annotation, by using ML algorithms for face detection and recognition.
One of the restrictions for the solution to be implemented is that the data set used for training should be small enough to avoid the need of manually annotation of the content for creating the ground truth. The proposed solution follows the concept of incremental learning and tests conducted showed that even with a small size dataset, the results obtained are satisfactory.
To facilitate the implementation of annotated images for the FotoInMotion tool, INESC has designed an innovative platform that other project partners can use to tag relevant objects and faces of previously specified individuals.
All Image Credits: INESC