In the vibrant and iconic city of New Orleans on June 19, 2022, winners of the third Shape Recovery from Partial Textured 3D Scans (SHARP) Workshop and Challenge were announced. This annual challenge is co-founded and run by global leader in 3D scanning technology Artec 3D, and the University of Luxembourg’s Interdisciplinary Centre for Security, Reliability and Trust (SnT) with the aim of advancing AI and machine learning with 3D data processing, and promoting 3D technology as an essential academic topic.
The challenge was announced in January this year, and the teams got started. In June, at the Computer Vision and Pattern Recognition Conference (CVPR) conference, three teams of finalists presented their work. The prizes – worth a total of €8,000 – were awarded by co-chairs Djamila Aouada (SnT) attending in person, and Kseniya ( , SnT) who attended virtually. CherenkovaArtec 3D
This annual challenge encourages the creation of breakthrough methods for recovering complete textured 3D scans from raw incomplete data through two challenges – Textured Partial Scan Completion and Sharp Edge Recovery.
The challenge: accuracy through algorithms
Through their tasks, teams had to accurately reconstruct a full 3D textured mesh from a partial 3D scan in the first challenge, and in the second – using a 3D object scan with smooth edges – reconstruct corresponding CAD model as a triangular mesh with sharp edges approximating the ground-truth sharp edges.
For the first challenge, participants were tasked with creating an algorithm superior to Artec 3D’s reference algorithm for two tracks: the first track, which required recovering textured human body scans from partial acquisitions, was won by the team from Dalian University of Technology, China. The second – recovering textured object scans from partial acquisitions, using a dataset which contained over 2,000 objects with different levels of complexity in texture and geometry – was won by ETH Zurich, Switzerland.
Breaking new ground in 3D data processing
Gleb Gusev, Chief Technology Officer of Artec 3D, said, “The AI/ML community have already made huge achievements in dealing with 1- and 2-dimensional data such as speech, text, and images using neural networks. But, working with 3D data is far more difficult – it’s a completely different beast and it’s still very new territory.”
Referring to the significance of this project and the possibilities AI can present, he continued, “By initiating and sponsoring these challenges, we are moving 3D data understanding to a new level, where we can accomplish far more, such as automatic data completion and automatic mesh-to-CAD conversion by using AI-based algorithms.”
“This is our third edition of the SHARP workshop and challenges, held once again in conjunction with a top-tier conference – CVPR. SnT and Artec3D have been working together on collecting, preparing and publicly sharing unique real 3D datasets that present high-complexities and diversities,” said Prof. Djamila Aouada of SnT. “We have designed these challenges in order to steer research advances towards realistic and complex tasks on 3D data.”
“The increasing interest in SHARP from the research community with 27 participants this year, and over 50 attendees during the live workshop, is a testimony to the relevance of SHARP.”
For more than five years, Luxembourg-based industry leader Artec 3D has worked in collaboration with SnT to conduct academic research projects and develop educational partnerships. Since its launch in 2009, SnT has continued to conduct internationally competitive research in information and communication technology with a goal of creating a significant socio-economic impact. During its collaboration with Artec 3D they have had research projects focused on body measurement research, educational collaborations, and security (3D face recognition).
Award presentation