Deep Learning-based Detection of Greek Locations and Landmarks: A Qualitative Analysis of Weakly Supervised Classification and Supervised Detection

The LoockMe platform integrates a wide array of artificial intelligence models that perform natural language processing and deep learning image analysis to assist audiovisual professionals in identifying the best locations for producing content. The latest iteration of the LoockMe dataset consists of images of Greek locations or landmarks with thousands of points of interest annotated and labeled by a diverse group of raters. Transfer learning was employed to adapt several YOLO models originally trained on the COCO dataset. The platform can label and annotate input images of Greek landscapes to identify suitable locations to film audiovisual content. The deep learning-based object detection model achieved high performance with a mAP score of 0.62. Additionally, the presented qualitative analysis sheds light on the real-world performance of the proposed detection model. The LoockMe artificial intelligence-powered platform aims to significantly accelerate location scouting for audiovisual production in Greece.