MetaFood CVPR 2024 Challenge on Physically Informed 3D Food Reconstruction: Methods and Results

越来越多的人对于计算机视觉在营养和饮食监测方面的应用感兴趣，这导致了对于食物的先进三维重建技术的发展。然而，高质量数据的稀缺和产学合作的限制制约了该领域的进展。基于最近三维重建技术的进步，我们主办了MetaFood研讨会及其物理信息三维食品重建挑战。这个挑战的重点是从二维图像中重建出食品物品的体积精确的三维模型，使用可见的棋盘格作为大小的参考。参与者的任务是为20个不同难度级别的食品物品重建三维模型，包括易、中、难三个级别。易级别提供了200张图片，中级别提供了30张图片，而难级别仅提供1张图片用于重建。总共有16个团队在最终测试阶段提交了结果。在这个挑战中开发的解决方案在三维食品重建方面取得了有希望的结果，具有显著的潜力来改善营养评估和营养监测中的食物份量估计。有关此研讨会挑战的更多详细信息和数据集访问，请访问https://sites.google.com/view/cvpr-metafood-2024。

Physically Informed 3D Food Reconstruction Challenge

Using advanced 3D reconstruction techniques to reconstruct volume-accurate 3D models of food items from 2D images, with a visible checkerboard as a size reference

Challenge focuses on reconstructing 3D models for 20 selected food items of varying difficulty levels, with 16 teams submitting results in the final testing phase. Promising results achieved with significant potential for improving portion estimation for dietary assessment and nutritional monitoring.

Recent advancements in 3D reconstruction for food items, limited collaboration between industry and academia.