Upstage OCR Achievements at ECCV 2022

Upstage proved its OCR technology by having 2 papers published at ECCV 2022. One paper is research on multimodal character recognition that reflects both linguistic and visual elements together when recognizing characters in photos, and the other is research on a learning strategy that adapts pretrained models to specific fields.

ECCV is regarded as one of the world's most prestigious computer vision conferences along with ICLR and CVPR, and was described as a top-tier conference with an acceptance rate of under 30%. According to the article, 5,150 papers were submitted to ECCV 2020, and only 1,360 papers (26%) were accepted.

Building on this achievement, Upstage plans to accelerate development of Upstage OCR Pack, which extracts desired information from images. It also plans to launch Upstage AI Pack, which allows users to leverage 3 AI technologies at once including OCR, recommendation, and semantic search, in the second half of this year.

• First paper: Achieved higher performance than existing vision-centric OCR through multimodal character recognition
• Second paper: Supplemented performance on unseen domain data through a custom learning strategy based on pretrained models
• Conference where the papers will be presented: ECCV 2022, October 23-27, Tel Aviv, Israel
• Upstage also highlighted its presentation track record at major conferences this year including NeurIPS, ICLR, CVPR, WWW