Within IDLab, AI for image and video data is approached from multiple angles: the transfer and compression of such data, the analysis of such data and the generation of it. IDLAB focuses on:
Specific research track differentiators include incorporating a mix of feature engineering and feature learning techniques, applying ML (such as GANs, CNNs, time series techniques) where needed, and the use of multimodal data and input, where object-mounted sensors (e.g., accelerometer on a bike) are combined with video data coming from multiple sources. Our work on detecting manipulated images and forensics has received attention in popular technical literature.
IDLAB Ghent has expertise in (applied) research on machine learning for image and video data, mainly (but not exclusively) within the domains of:
Healthcare
e.g. heart-volume extraction from video, medical image analysis for skin cancer (BARCO collaboration) and ophthalmology
Industry 5.0
e.g. automated quality control (e.g., weld analysis) and condition monitoring
(XR) Media generation and enrichment
e.g. logical story unit detection in video, article segmentation in newspaper archives, automatic geolocalization, sports analytics
Media forensics – battling disinformation
e.g. detection and localization of manipulated parts of images
