A research team, led by Professor Sungryul Baek from the Graduate School of Artificial Intelligence at UNIST has achieved remarkable success at the ‘ICCV 2023 Workshop on Observing and Understanding Hands in Action (HANDS),’ which took place in Paris, France from October 2 to 6, 2023. The competition aimed to assess the accuracy of computer vision models in interpreting the shapes of moving hands and objects.
Under the leadership of Professor Baek, researchers JiHyeon Kim, HoSeong Cho, JunUk Cha, and Elkhan Ismayilzada from his lab collaborated to participate in the challenge. Their outstanding efforts led them to secure the third prize by presenting a groundbreaking model, capable of predicting the 3D mesh of two hands and the pose of an object solely from a single RGB image, without the need for additional post-processing. A 3D mesh refers to a polyhedron used to represent the shape of an object in a three-dimensional space, comprising a collection of polygons and vertices.
The research team emphasized that their model stood out for its ability to accurately predict the interaction between both hands and objects, unlike previous models that could only anticipate one-handed or two-handed poses. This innovative approach garnered significant recognition from the judging panel.
Expressing his thoughts on the achievement, Professor Baek remarked, “The restoration of mesh on two hands interacting with objects is a crucial field with immense potential for various applications, including augmentation, virtual reality (AR/VR), and the metaverse.” He further added, “We remain committed to pursuing excellence and delivering promising outcomes in the future.”
Meanwhile, the International Conference on Computer Vision (ICCV) is a research conference sponsored by the Institute of Electrical and Electronics Engineers (IEEE), held every other year. It is considered to be one of the top conferences in computer vision, alongside CVPR and ECCV. The ICCV 2023 took place online