Reproduce “Advancing Food Nutrition Estimation via Visual-Ingredient Feature Fusion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Huiyan Qi, Bin Zhu, Chong-Wah Ngo, Jingjing Chen, Ee-Peng Lim
This research introduces the FastFood dataset, a large-scale collection of 84,446 images with detailed nutritional and ingredient annotations for fast food items. It also presents the Visual-Ingredient Feature Fusion (VIF²) method, which systematically incorporates ingredient information alongside visual features, reducing caloric Mean Absolute Error (MAE) by 48.1% on the new dataset compared to visual-only approaches.
Key insights
- Nutritional values are intrinsically linked to a food item's constituent ingredients, not solely its visual appearance.
- Large-scale datasets with comprehensive ingredient and nutritional annotations are essential for accurate food nutrition estimation.
- Effectively fusing visual and textual ingredient features, even with predicted ingredients, leads to substantial improvements in nutritional prediction accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.