The 2021 edition of the DocVQA challenge, which is a long term challenge in ICDAR has come to an end. The winners will be presenting their solutions at the workshop.
By Document Visual Question Answering ( DocVQA ) we refer to a generic paradigm for purpose-driven document analysis and recognition, where natural language questions drive the information extraction and document understanding processes. Over and above document images, the workshop aims to highlight the role of image-rendered textual information in other VQA branches as well, such as Scene-Text VQA and data visualization question-answering ( DVQA ).
Organizing the DocVQA workshop at ICDAR 2021 is a natural step after the success of the Document Visual Question Answering (DocVQA) challenge organized as part of “Text and Documents in the Deep Learning Era” Workshop in CVPR 2020. This year we have extended and brought the DocVQA challenge to ICDAR 2021 as a long-term ICDAR challenge , and we aspire that the DocVQA workshop we propose here will provide a space for discussion both on the challenge itself, and on the importance of reading systems for VQA in general.
Half day workshop during September 5 – 7, 2020. Exact date and time will be updated later.
The workshop will be a half day workshop. The program includes invited talks from experts in the Document understanding space, a panel discussion and a session on the DocVQA challenge.
2021 edition of DocVQA challenge is organized as a long-term challenge of ICDAR 2021. The challenge comprises two tasks:
Continuation of Task 2 from 2020 edition of the challenge. This task deals with QA on a collection of document images with same template.
Introduction of a new Task3, which deals with VQA on infographics