Iconographic Image Captioning for Artworks

Image captioning implies automatically generating textual descriptions of images based only on the visual input. Although this has been an extensively addressed research topic in recent years, not many contributions have made domain art historical data. In particular context, task image is confronted with various challenges such as lack large-scale datasets image-text pairs, complexity meaning associated describing artworks and need for expert-level annotations. This work aims to address some those by utilizing a novel dataset artwork annotated concepts from Iconclass classification system designed iconography. The annotations are processed into clean description create suitable training deep neural network model task. Motivated state-of-the-art results achieved captions natural images, transformer-based vision-language pre-trained fine-tuned using dataset. Quantitative evaluation performed standard metrics. quality generated model’s capacity generalize new data explored employing collection paintings performing analysis relation between commonly artistic genre. overall suggest that can generate meaningful exhibit stronger relevance particularly comparison obtained models trained datasets.