Automatic Analysis of Amateur Film and Video Collections

Amateur film and video collections are often only sparsely documented and the contents of the material are only vaguely known. The progress in deep learning based computer vision methods over the last decade provides an opportunity to automatically generate additional descriptions that would help exploring and finding content, even if they do not meet the level of professional archive documentation. While the automatically generated metadata can facilitate interactive search and exploration, it will always need to be used with a human user in the loop, as it will not be perfect. In this paper, we describe a framework including visual content analysis tools and a web-based visualisation of the extracted metadata.