Recognising people in videos using a pre-trained deep learning model

In this talk, I will present how to recognise speakers and their speaking sessions in a filmed meeting using deep learning (deep convolutional neural networks; CNN). This project uses the "transfer learning" approach. Pre-trained CNN models, such as the VGG face descriptor used in this project, enable everyone to analyse photos or videos without training his own CNN. I will explain how to use a pre-trained model to extract face features and use clustering methods to identify different people without knowing their identity in advance. Results of a real case will be shown. Application and restriction of this method will be discussed.