Multimodal interpretability

Visuo-Linguistic Grounding

Process of anchoring linguistic symbols (words, phrases) to concrete entities or concepts in visual data, fundamental for model explanations linking text and image to be semantically correct and understandable.

← Wstecz