UNIST Professor Baek Seung-ryeol's Team Wins 1st Place in European Conference on Computer Vision (ECCV) Competition

Award in ARCTIC Category at Hand Competition

AI Technology Reducing Distortion by 3 Times in Video

The research team led by Professor Baek Seung-ryul at the Graduate School of Artificial Intelligence, UNIST (President Park Jong-rae), topped the Hands workshop challenge at the European Conference on Computer Vision (ECCV) with AI video technology that reduces distortion.

Professor Baek Seung-ryeol, Graduate School of Artificial Intelligence, UNIST.

The challenge was held as a workshop event at the 2024 European Conference on Computer Vision on the 30th of last month. It was divided into four categories, and Professor Baek's team won first place in the ARCTIC category.

The ARCTIC category evaluates the accuracy of computer vision models that reconstruct 3D meshes from videos capturing the movements of both hands and objects.

The task requires reconstructing the mesh using only third-person camera information without using 3D scanned template data. A mesh refers to data reconstructed as a polyhedron representing the object.

Professor Baek's team (students On Jeong-wan, Kwak Kyung-hwan, Kang Geun-young, Hwang Hye-in, Hwang Su-hyun, Cha Jun-wook, and Han Jae-wook) was awarded for presenting 3D reconstruction technology that is free from distortion even from side views.

Existing technologies appear to produce natural 3D reconstructions from the camera viewpoint but exhibit distortions in the distance and contact areas between hands and objects when viewed from the side. In contrast, Professor Baek's team applied the latest ‘3D Gaussian Splatting technology’ to reduce the distance error between hands and objects by about three times (70 cm).

Professor Baek said, “Since people's hands constantly interact with objects in real life, accurate 3D mesh reconstruction is important,” adding, “This can contribute to creating vivid virtual reality and metaverse experiences.”