Description and task

In this competition you will need to build a Visual Question Answering model over document images.

Shared task focuses on non-extractive questions. Unlike more traditional setting where each answer is just a span within the document text, answers to the GenDocVQA-2024 questions have to be generated by the VQA model. In addition, it is necessary to comply with the robustness requirements — the solution must demonstrate effectiveness dealing with various question types and question paraphrasing.

The language of the documents and questions in GenDocVQA-2024 is English.

Timeline

01.06.2024 — start of the shared task;

01.09.2024 — deadline for paper submissions.