- Task 1 - Single Page Document VQA
- Task 2 - Document Collection VQA
- Task 3 - Infographics VQA
- Task 4 - MP-DocVQA
method: (Baseline) Database2021-04-12
Authors: DocVQA Organizers: R. Tito, D. Karatzas, E. Valveny
Affiliation: CVC-UAB
Description: Two-step method that uses extracted key-value pairs from Amazon Textract to build a database-like data structure to rank and answer the questions that are manually parsed to SQL language.
Please check cited paper for detailed information.
method: (Baseline) Text spotting - BERT2021-04-12
Authors: DocVQA Organizers: R. Tito, D. Karatzas, E. Valveny
Affiliation: CVC-UAB
Description: Two-step method which first ranks the documents according to the relevance between the document and the question and then uses extractive BERT QA method to get the answer from the relevant documents.
Please check cited paper for detailed information.
Description Paper Source Code
| Date | Method | ANLSL | Retrieval MAP | |||
|---|---|---|---|---|---|---|
| 2021-04-12 | (Baseline) Database | 0.7068 | 71.06% | |||
| 2021-04-12 | (Baseline) Text spotting - BERT | 0.4513 | 72.84% |