Automatic Image Captioning - Einzelansicht

In the project, we will take a look at the state of the art in automatic image caption generation with deep neural fitting and coherent sentences. Generating image captions is a prime multimodal learning task, which connects computer vision and natural language processing. Almost all image captioning models adopt the encoder-decoder framework with a visual attention mechanism. The encoder encodes input images into fixed-length vector features, and the decoder decodes image features into descriptions word by word. Based on a survey of image caption approaches with available source code, our goal is to deploy the most promising image captioning approaches onto our GPU cluster and evaluate their performance using various benchmark datasets. The best performing approach is intended to be used in an upcoming digital humanities research project on the analysis of image feed curation algorithms in social networks.

Time and place will be announced at the project fair.

Abschlusspräsentation und Ausarbeitung

B.Sc. Medieninformatik / Informatik

M.Sc. Computer Science for Digital Media

M.Sc. Human-Computer Interaction

M.Sc. Digital Engineering