Continuous 3D Food Reconstruction: Team PerBite Wins the MetaFood CVPR 2026 Challenge

Continuous 3D food reconstruction: LogMeal’s Team PerBite wins the MetaFood CVPR 2026 challenge. The 2026 continuous 3D food reconstruction challenge was held on 3 June 2026 in Denver. In fact, the competition pushed the boundaries of food computer vision. Participants had to reconstruct the evolving 3D shape and volume of food from egocentric eating videos. Moreover, these sequences involve continuous rotation, deformation, occlusion and breakage. As a result, participants must infer geometry and temporal coherence even when food leaves the field of view. Ultimately, this requirement pushes 3D vision systems toward truly robust modeling.

PerBite wins the continuous 3D food reconstruction challenge

LogMeal proudly announces that Team PerBite clinched first place in this continuous 3D food reconstruction challenge. Additionally, the official results include Ahmad AlMughrabi, Farid Al‑Areqi, David Fernández Gómez, Umair Haroon, Marc Bolaños, Ricardo Marques and Petia Radeva among the team members. Furthermore, the team excelled in both phases of the evaluation. They combined accurate volume estimation with precise 3D shape reconstruction under rigorous conditions. Moreover, Marc Bolaños and Petia Radeva, co‑founders of LogMeal, played leading roles. Their participation and the team’s victory highlight LogMeal’s deep expertise in food computer vision. Our success also shows our ability to translate research into practical solutions.

Why continuous 3D food reconstruction matters

Continuous 3D reconstruction from monocular video is considered a frontier problem in computer vision. However, real‑world eating scenarios include occlusions from hands and utensils and deformation as food is cut or bitten. There is also no explicit scale reference. The 2026 edition built on the previous multi‑food reconstruction task and emphasised temporally coherent reconstruction of food during consumption. The MetaFood challenge evaluates methods under dynamic, non‑rigid conditions. As a result, it advances dynamic scene reconstruction, object pose estimation and physics‑aware modelling.

Moreover, accurate 3D reconstruction and portion estimation are core technologies for automated nutritional assessment and dietary tracking. In addition, they underpin real‑life food intelligence. These capabilities allow us to move beyond single‑image recognition and to understand how much food is consumed and how it changes over time. Consequently, this information is key for health, sports, foodservice and digital nutrition platforms.

Looking ahead for 3D food AI

Team PerBite’s victory is a proud milestone for our scientific and technical team. It is also a strong validation of our approach to food AI. The win reinforces LogMeal’s leadership in food recognition, continuous 3D reconstruction and portion estimation. It demonstrates that our research can solve complex real‑world problems.

We will continue to collaborate with the academic community. Furthermore, we will participate in initiatives like MetaFood. Therefore, we remain committed to bringing the future of food computer vision into practical applications. If your organization is exploring nutrition technology, dietary assessment or AI‑powered food services, we invite you to get in touch. Additionally, contact us to discover how LogMeal’s solutions can help.

or contact us today to learn how we can help streamline your practice.