Personal Identification Based Automatic Face Annotation in Multi-View Systems
Plamen Hristov, Ognian L. Boumbarov, Krasimir Vachev · 2022
We propose a method for automatic face annotation in a closed multi-view environment. In such an environment faces are automatically detected and images are collected from several sources. Then, for the purpose of real-time labelling, a model is incrementally trained with every new person who enters the environment. The incremental learning model is validated on the Pandora dataset, which is split into different classes for each incremental step. Active learning is used to determine the type of images needed for training based on a predefined budget.