OmniConsistency is a multimodal benchmark to evaluate cross-modal consistency in generated image-text pairs. It consists of 22 distinct visual styles (e.g., "Ghibli", "LEGO", "Van Gogh") and contains human-annotated captions describing synthetic artworks.
This is the Turkish-translated version of the OmniConsistency dataset, originally released by Showlab. It includes 22 stylistic splits with multilingual captions, where all English… See the full description on the dataset page: https://huggingface.co/datasets/salihfurkaan/OmniConsistency-TR.
