Deep Tagging

Deep learning for automatic image tagging

The goal of the project is to develop automatic image tagging methods. In a first step, we setup a large-scale dataset of images stemming from different semantic categories (e.g. product images, logos, indoor recordings, outdoor photos etc.). Next, we investigate methods based on deep learning that are able to learn the characteristics of the different semantic categories. These methods are trained on our dataset and will be used in a subsequent step to predict semantic labels (tags) for new images automatically. The basis for our investigation are pre-trained neural networks. To perform the necessary domain adaption we retrain the nets and adapt their architecture appropriately.
 

Project partner: mediamid digital services GmbH

Peter Alexander Kopciak

Project partners