Implicit-Scale 3D Reconstruction from Monocular Multi-Food Images
收藏资源简介:
该数据集由普渡大学等机构联合构建,旨在通过单目图像实现多食物场景的隐式尺度三维重建。包含10个精心设计的餐饮场景,共24个三维物体,数据来源于高精度三维扫描仪采集的MetaFood3D对象集。通过移除显式物理参照物,要求算法从餐具等上下文信息推断尺度,真实模拟了现实餐饮场景的遮挡和复杂空间布局。主要应用于膳食评估领域,为解决食物体积估算中的几何推理和尺度模糊问题提供基准。
This dataset was co-developed by Purdue University and other institutions, aiming to accomplish implicit-scale 3D reconstruction of multi-food scenes using monocular images. It contains 10 meticulously designed dining scenarios with a total of 24 3D objects, with data sourced from the MetaFood3D object collection acquired via high-precision 3D scanners. By eliminating explicit physical reference objects, it mandates algorithms to infer scale from contextual cues such as tableware, and realistically mimics the occlusion and complex spatial layouts of real-world dining scenarios. Primarily applied in the field of dietary assessment, this dataset serves as a benchmark for addressing geometric reasoning and scale ambiguity problems in food volume estimation.



