This is an XML dataset of 17 lecture recordings randomly sampled from the lectures recorded at the Faculty of Informatics, Brno, Czechia during 2010–2016. We drew a stratified sample of up to 25 video frames from each recording. In each video frame, we annotated lit projection screens and their condition. For each lit projection screen, we annotated lecture materials shown in the screen. The dataset contains 699 projection screen annotations, and 925 lecture materials. Video frames and lecture materials are encoded as 8-bit PNG images.
The XML schema schema.xsd describes the structure of the dataset. Section
3 of [Novotny18] describes the details of building the dataset. The linked
presentation
outlines the system for
which the dataset was built. In our experiments,
we split the dataset into 16 training videos and one validation video using
leave-one-our cross-validation.
| [Novotny18] | NOVOTNÝ, Vít. Slide retrieval based on lecture recordings. Brno, 2018. 27 pp. Available from: https://github.com/video699/research-report/releases/download/2018/01/19/handout.pdf |