▲ KCCV2023 (Korean Conference on Computer Vision 2023)

10th KCCV held, domestic vision research ↑ Rapid growth

One place for domestic and international computer vision research presentations, with many corporate sessions

As the era of artificial intelligence dawns, existing computer vision technology is rapidly developing and greatly expanding its market area. As it is a technology in the spotlight, interest in the domestic computer vision society has also increased, and this year, the conference hall was packed with attendees, recording the highest participation rate ever.

The Korean Computer Vision Society held the KCCV 2023 (Korean Conference on Computer Vision 2023) at the COEX Grand Ballroom in Samseong-dong, Seoul on the 7th.

KCCV, now in its 10th year, has made remarkable progress compared to the first event held in 2014. Currently, there are about 300 Korean papers published in the world's top academic conferences, including CVPR and ECCV, and KCCV is also growing in size thanks to the development of domestic vision research.

The number of pre-registrants reached 1,100 and the number of sponsoring companies reached 20, breaking the record. This event, in which various industries and research institutes such as Qualcomm, POSCO DX, Kakao Brain, Korea Electronics Technology Institute, and Korea Institute of Science and Technology participated, consisted of sessions with papers presented at world-renowned academic conferences in the field of computer vision such as ECCV 2022, NeurIPS 2022, and CVPR 2023.

Consisting of a main conference, workshops, and tutorials, KCCV 2023 will be held over four days and will consist of △oral presentations △Doctoral Colloquium △corporate presentations △poster and demo sessions △invited lectures △panel discussions, etc. On the last day, workshops and programming practice tutorials will be held to provide an opportunity to directly practice the latest papers.

▲Lee Jun-ho, President of the Korean Computer Vision Society, giving a congratulatory speech

The following papers were presented on the first day: △ Realistic Blur Synthesis for Learning Image Deblurring by Professor Jo Seong-hyeon of Pohang University of Science and Technology, △ B-Spline Texture Coefficients Estimator for Screen Content Image Super-Resolution by Professor Jin Kyeong-hwan of Daegu Gyeongbuk Institute of Science and Technology, and △ Sound-guided Semantic Video Generation by Professor Kim Sang-pil of Korea University.

Among these, Professor Cho Seong-Hyeon's presentation was a study presented at the European Conference on Computer Vision (ECCV) 2022, in which he conducted research to improve the virtual blur data set used for learning existing deblurring techniques into a more realistic data set.

 

In order to sharpen the image, a technology is needed to remove the motion blur of the object. This deblurring technology is adopted in various applications, and in order to improve the function, a learning data set is needed.

Existing techniques have the limitation of being difficult to obtain a blur data set, so they create a synthetic data set and input it into learning. However, due to the limitation of 'homology' that is identical or similar between synthetic data, it was necessary to secure a more realistic synthetic data set. Professor Cho conducted research to improve this problem.

Professor Jin Kyung-hwan presented a new algorithm that matches the resolution differences between images captured by a camera and displays in a study presented at CVPR2023. Professor Kim Sang-pil expanded on the paper presented at CVPR2022 and presented advanced research results called a voice-based image generation model.

▲AI talent recruitment consultation held at the KCCV2023 exhibition booth

In addition, companies such as Qualcomm, POSCO DX, and Recon Labs participated in the corporate session and introduced technologies being researched in industrial fields and the latest products being introduced to the market. POSCO DX explained computer vision technology and cases utilized in manufacturing sites, and Qualcomm introduced its autonomous driving platform, the Snapdragon Ride Platform.

Recon Labs introduced 3Dpresso, a SaaS service that models subjects in 2D photos, images, and videos into 3D. Yoon Kyung-won, CTO of Recon Labs, emphasized 'time to market' in the development front line responding to the rapidly changing market, saying, "Requirements and response methods differ for each AI model, and the market and technology updates are so fast that it can be difficult to keep up with papers."

Lee Jun-ho, the president of the Korean Computer Vision Society, was proud of the event, saying, “Through the KCCV event, excellent researchers were able to spread world-class research results and research methods, contributing to the development of the domestic computer vision field.” He also said, “I hope that KCCV will become a venue for domestic researchers from academia, industry, and research in the field of computer vision to communicate closely and have active discussions.”

Choi Jong-Hyeon, the chair of the KCCV2023 program and professor at Yonsei University, said, “Research presented at KCCV is now achieving qualitative development by winning various awards, such as the Test of Time Award and Outstanding Paper Award, at world-class academic societies.” He added, “As the demand for computer vision technology in industry and the market increases, the attention and status of the Society for Computer Vision are also increasing.”

▲A long line of visitors waiting to attend KCCV 2023


▲Poster session