has been designed around providing deep-learning solutions. The main focus currently revolves around image recognition. Current training methods primarily make use of a method called transfer learning; retraining pre-trained networks with a new dataset. Future projects aim to train new networks from scratch to allow greater flexibility in the weighting parameters.
We aim to be as tranparent as realistically possible throughout the development and updates of our projects. Images used for trraining purposes either come from the ImageNet database, or our own dataset of images with consent to use said images given from the subject of the photo. For the purposes of our main current project, various images of the devlopment staff have been used.
-
Our development is primarily based in MATLAB, making use of the Deep Learning Toolbox
-
Transfer learning for out projects is mainly done using either the ResNet 101 or ResFace (which is based on ResNet) network models