Enregistré dans:
Détails bibliographiques
Auteurs principaux: Bright, Darrin, Raj, Rakshith, Keisham, Kanchan
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2512.22304
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866909976706940928
author Bright, Darrin
Raj, Rakshith
Keisham, Kanchan
author_facet Bright, Darrin
Raj, Rakshith
Keisham, Kanchan
contents Accurate food nutrition estimation from single images is challenging due to the loss of 3D information. While depth-based methods provide reliable geometry, they remain inaccessible on most smartphones because of depth-sensor requirements. To overcome this challenge, we propose PortionNet, a novel cross-modal knowledge distillation framework that learns geometric features from point clouds during training while requiring only RGB images at inference. Our approach employs a dual-mode training strategy where a lightweight adapter network mimics point cloud representations, enabling pseudo-3D reasoning without any specialized hardware requirements. PortionNet achieves state-of-the-art performance on MetaFood3D, outperforming all previous methods in both volume and energy estimation. Cross-dataset evaluation on SimpleFood45 further demonstrates strong generalization in energy estimation.
format Preprint
id arxiv_https___arxiv_org_abs_2512_22304
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PortionNet: Distilling 3D Geometric Knowledge for Food Nutrition Estimation
Bright, Darrin
Raj, Rakshith
Keisham, Kanchan
Computer Vision and Pattern Recognition
Accurate food nutrition estimation from single images is challenging due to the loss of 3D information. While depth-based methods provide reliable geometry, they remain inaccessible on most smartphones because of depth-sensor requirements. To overcome this challenge, we propose PortionNet, a novel cross-modal knowledge distillation framework that learns geometric features from point clouds during training while requiring only RGB images at inference. Our approach employs a dual-mode training strategy where a lightweight adapter network mimics point cloud representations, enabling pseudo-3D reasoning without any specialized hardware requirements. PortionNet achieves state-of-the-art performance on MetaFood3D, outperforming all previous methods in both volume and energy estimation. Cross-dataset evaluation on SimpleFood45 further demonstrates strong generalization in energy estimation.
title PortionNet: Distilling 3D Geometric Knowledge for Food Nutrition Estimation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.22304