Overview - Document Visual Question Answering

The "Document Visual Question Answering" (DocVQA) challenge, focuses on a specific type of Visual Question Answering task, where visually understanding the information on a document image is necessary in order to provide an answer. This goes over and above passing a document image through OCR, and involves understanding all types of information conveyed by a document. Textual content (handwritten or typewritten), non-textual elements (marks, tick boxes, separators, diagrams), layout (page structure, forms, tables), and style (font, colours, highlighting), to mention just a few, are pieces of information that can be potentially necessary for responding to the question at hand.

The DocVQA challenge is a continuous effort linked to various events. The challenge was originally organised in the context of the CVPR 2020 Workshop on Text and Documents in the Deep Learning Era. From this first event a paper with the results was presented in the Document Analysis Systems International Workshop that can be found here. The second edition will take place in the context of the Int. Conference on Document Analysis and Recognition (ICDAR) 2021.

What is the issue at the top of the pyramid? Retailer calls/ other issues Which is the least critical issue for live rep support? Retailer calls/other issues Which is the most critical issue for live rep support? Product quality/liability issues		​​​How many females are affected by diabetes 3.6% What percentage of cases can be prevented 60% What could lead to blindness or stroke diabetes

Figure 1. Example documents from DocVQA Task 1 (left) and Task 3 (right) with its Questions and Answers.

 

Contemporary Document Analysis and Recognition (DAR) research tends to focus on generic information extraction tasks (character recognition, table extraction, word spotting), largely disconnected from the final purpose the extracted information is used for. The DocVQA challenge, seeks to inspire a “purpose-driven” point of view in Document Analysis and Recognition research, where the document content is extracted and used to respond to high-level tasks defined by the human consumers of this information. In this sense DocVQA provides a high-level task that should dynamically drive information extraction algorithms to conditionally interpret document images.

On the other hand, Visual Question Answering (VQA) as it is currently applied in real scene images is vulnerable to learning coincidental correlations in the data without forming a deeper understanding of the scene. In the case of DocVQA, more profound relations between the question aims (as expressed in natural language), and the document image content (that needs to be extracted and understood) are necessary to establish.

A large-scale dataset of document images reflecting real-world document variety, along with question and answer pairs will be released according to the schedule on the right.

The challenge comprises three different tasks that we briefly describe here. For more detailed information refer to Tasks section.

• Task 1 - Single Document VQA: Is a typical VQA style task, where natural language questions are defined over single documents, and an answer needs to be generated by interpreting the document image. No list of pre-defined responses will be given, hence the problem cannot be easily treated as an n-way classification task.
  • UPDATE: This task remains open for the community to keep working on, but it won't be part of the new ICDAR2021 competition.
  • If you use dataset for Task1, consider citing

@InProceedings{docvqa_wacv,
    author    = {Mathew, Minesh and Karatzas, Dimosthenis and Jawahar, C.V.},
    title     = {DocVQA: A Dataset for VQA on Document Images},
    booktitle = {WACV},
    year      = {2021},
    pages     = {2200-2209}
}

        

• Task 2 - Document Collection VQA: Is a retrieval-style task where given a question, the aim is to identify and retrieve all the documents in a large document collection that are relevant to answering this question as well as the answer.
  • UPDATE: This task has been redesigned for the ICDAR2021 edition to take into account both the evidences provided and the question answering performance of participating methods.

 

• Task 3 - Infographics VQA (NEW!): This is a new task , introduced as part of the 2021 challenge ( the ICDAR 2021 edition). The task is similar to the task1. Only difference is that images are infographics.  More details of the task can be found under "Tasks" tab.

 

 

References

[1] Minesh Mathew, Dimosthenis Karatzas, C. V. Jawahar, "DocVQA: A Dataset for VQA on Document Images", arXiv:2007.00398 [cs.CV], WACV 2021

[2] Minesh Mathew, Ruben Tito, Dimosthenis Karatzas, R. Manmatha, C.V. Jawahar, "Document Visual Question Answering Challenge 2020", arXiv:2008.08899 [cs.CV], DAS 2020