Couger Wins Top 10 ECCV Challenge, One of the World's Three Largest International Conferences: Unique Method for Image Filter Removal Maximizes Learning Effectiveness

Couger Inc. (Headquarters: Shibuya-ku, Tokyo; CEO: Atsushi Ishii) is pleased to announce that it has been ranked 9th in the ECCV 2022 competition, one of the world's three largest international conferences in the field of computer vision.

In 2022, a lot of attention was paid to the high application capabilities of AI, such as GPT3's interactive AI and Stable Diffusion's image generation AI, which was trained on a dataset of "only beautiful images" called LAION-Aesthetics. Breakthroughs in AI technology depend heavily on the quality of training data. In our daily lives, the quality of training data also plays a fundamental role in the underlying foundations of face recognition used in smartphones and PCs, object recognition used in automated driving, and other applications.

On the other hand, while the processing filters used in many social networking images and the increasing sophistication of cameras are convenient, they have a negative impact on machine learning in that they produce images that are different from the real image. Couger's image recognition team participated in ECCV's Instagram Filter Removal Challenge to develop a unique method for removing filters from processed Instagram images. In the end, Couger's team was selected as a finalist and placed 9th out of 114 participants.

Applying methods adopted in the past at major international conferences

Couger's image recognition team applied U-Net, a leading deep learning method of Semantic Segmentation, to achieve the goal of this competition: to generate realistic and visually plausible images that remove the influence of image processing filters as much as possible. Couger has achieved high accuracy with U-Net-based solutions at major international conferences in the past, has won several competition awards and has had papers accepted for publication.

*Each participating team's solution is published below

https://arxiv.org/pdf/2210.08997.pdf

Couger is focusing on research and development that will lead to the practical application of Human-like AI, including the installation of “RACHEL”, a Human-like AI, in 5,000 FamilyMart stores. The use of image and video data on Instagram, TikTok, and other digital spaces will continue to expand, and the use of data in stores using cameras and sensors is certain to expand as well. Couger will continue to utilize a variety of AI technologies, including image recognition such as gaze recognition and facial recognition, as well as, natural language processing.

About ECCV

ECCV (European Conference on Computer Vision) is considered one of the three major international conferences in the field of computer vision (CVPR, ICCV). The 2022 edition of this conference was held in October.

Adoption of Past International Conferences

https://couger.co.jp/news-en/2021/08/23/cvpr2021/

https://couger.co.jp/news-en/2021/11/29/iccv2021/

● Recruiting Information
Couger supports autonomous and bold challenges, and aims to be a place where each individual can pursue their own unique skills and creativity. Currently, we are actively recruiting Project Manager, Server Application Engineer, Front End Engineer and QA Engineer. We also offer casual meetings for those who would like to just hear about us before applying, so please feel free to contact us!

Click here for a list of available positions and the casual meeting application form.