CMC reaches world top 12 in text recognition

The CATI-VLM (Visual Document Understanding) model developed by CMC Technology Application Institute (CMC ATI) has surpassed many international competitors to reach the top 12 in the world and top 1 in Vietnam in the rankings recently announced by Robust Reading Competition (RRC) in June 2025 in the Document Visual Question Answering (DocVQA) category.

RRC ranking in DocVQA category 6/2025.

In the context of digital transformation and artificial intelligence application transformation in Vietnam taking place strongly, OCR technology (Optical Character Recognition) plays an increasingly important role in digitizing documents, automating business processes, saving costs and improving management efficiency. However, with the characteristics of Vietnamese with accents and handwriting, the recognition problem does not stop at 'reading words', but requires the model to have the ability to understand the context comprehensively.

Recently, CMC Technology Application Institute (CMC ATI) announced the CATI-VLM (Visual Document Understanding) model - developed by the research team from a 5TB large data warehouse, surpassing many international competitors to reach the top 12 in the world and top 1 in Vietnam in the rankings just announced by Robust Reading Competition (RRC) in June 2025 in the Document Visual Question Answering (DocVQA) category.

Robust Reading Competition (RRC) is a prestigious scientific playground, (https://rrc.cvc.uab.es/) organized by the Computer Vision Center (CVC) of the Universitat Autònoma de Barcelona (UAB) Spain, a prestigious research facility in the world in the field of computer vision.

The competition was initiated in 2011 and is held annually within the framework of the International Conference on Text Analysis and Recognition (ICDAR) - one of the world's leading forums in the field of computer vision. The competition attracts a large number of researchers and engineers from universities, research institutes and large technology corporations such as Tsinghua University, Hyundai Motor Group, Tencent... RRC's problems are designed to promote technological progress, closely linked to practical problems from translation, enterprise data management to urban analysis and historical document processing.

Dr. Dang Minh Tuan, Director of CMC ATI shared: "We are very pleased that the research capacity of the CMC team has been affirmed through a prestigious global playground like RRC. In just a short time, the research team has achieved high rankings, demonstrating its international competitiveness with big names from developed countries. More importantly, this is a clear demonstration of the ability to master technology to solve specific problems of the Vietnamese language and specialized fields in Vietnam."

Dr. Dang Minh Tuan, Director of CMC ATI.

CATI-VLM differs from traditional OCR in that it not only extracts characters, but also understands multiple layers of information: text content, non-text elements (tick boxes, checkboxes, charts, signatures, formulas), layout (page structure, tables, forms) and style (fonts, highlights, etc.). The model can answer visual questions posed on document images, similar to ChatGPT, without having to learn specific forms beforehand.

Notably, on the RRC rankings, CATI-VLM with only 3 billion parameters achieved the highest accuracy in 4/7 datasets, surpassing many Big Tech models such as Deepseek (27 billion parameters), GPT-4 Vision Turbo + Amazon Textract OCR (top 34) or Baidu (top 22).

The achievement also shows a practical approach, focusing on mastering core technology, optimizing the model to suit Vietnam's infrastructure conditions instead of chasing parameter scale.

Sample College Admission Application Form

The text has been recognized from the handwriting in the image above.

Mr. Nguyen Trung Chinh, Chairman of the Board of Directors, Executive Chairman of CMC Technology Group, emphasized: "This is the result of more than a decade of persistent investment in technology research and development (R&D). CMC's high achievements in the international technology playground affirm the strategy of mastering Vietnamese technology, coupled with the orientation of AI Transformation and entering the global market. We believe that Vietnamese intelligence is fully capable of standing shoulder to shoulder with global Big Tech, creating a worthy position on the world technology map."

CATI-VLM will be applied in the product chain of the C.OpenAI ecosystem, including: CLS virtual assistant for reviewing legal documents, CMC SmartDoc - digital document conversion platform, CMC KMS knowledge management system, automatic reporting system for smart offices and new generation Agentic Documents applications.

QUANG HUY

Source: https://nhandan.vn/cmc-dat-top-12-the-gioi-ve-nhan-dang-van-ban-post891252.html