Nutrition5k: RGB-D Food Imagery with Per-Ingredient Mass, Calories and Macronutrients (Google Research)
- Provider
- Thames, Karpur, Norris, Xia, Panait, Weyand and Sim (Google Research)
- License
- CC-BY-4.0 (source)
- Stored at
- Google Cloud Storage, gs://nutrition5k_dataset/nutrition5k_dataset (us)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Health & Biomedical
- Source
- https://arxiv.org/abs/2103.03375
About this dataset
Visual and nutritional data for about 5,000 real cafeteria plates scanned in Google cafeterias in California, released with Thames et al., 'Nutrition5k: Towards Automatic Nutritional Understanding of Generic Food' (CVPR 2021). Each of 5,006 dishes was captured on a custom rig with four rotating side-angle videos from Raspberry Pi cameras at 30 and 60 degree angles, plus an overhead RealSense RGB-D image when available (raw 16-bit depth at 10,000 units per meter, and colorized depth). Dishes were built incrementally, one ingredient at a time, and weighed, so labels give a fine-grained ingredient list, per-ingredient mass, total mass and calories, and fat, protein and carbohydrate masses, with per-gram nutrition taken from the USDA Food and Nutrient Database. Listed 2026-11-19: nutrition5k_dataset/ holds imagery/ (side_angles videos, realsense_overhead), metadata/ (dish_metadata_cafe1.csv, dish_metadata_cafe2.csv, ingredient_metadata.csv), dish_ids/ (official train and test splits) and scripts/ (frame extraction and evaluation of calorie and macronutrient regression). A 181 GB tarball of everything sits at the bucket root. Released under CC BY 4.0; cite Thames et al. 2021.
How to access this dataset in Python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('nutrition5k_dataset/nutrition5k_dataset/', detail=False)[:10]Lists the files with anonymous access. Runs on any machine, with no cloud account.
Data structure
Files
Example file: nutrition5k_dataset/dish_ids/splits/depth_test_ids.txt
Analyze this dataset without downloading it
On Scigantic, Nutrition5k: RGB-D Food Imagery with Per-Ingredient Mass, Calories and Macronutrients (Google Research) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Health & Biomedical datasets
- TIGER Training
- WSSS4LUAD: lung adenocarcinoma H&E patches for weakly supervised tissue segmentation (tumor, stroma, normal)
- IDC QIN PET Phantom: Phantom PET (2 patients)
- COph100: infant retinal fundus images for retinopathy of prematurity (ROP) registration, with vessel masks and control points
- IDC VAREPOP-APOLLO: Esophageal Carcinoma, Head and Neck Squamous Cell Carcinoma, Lung Adenocarcinoma and 5 more CT, MRI, PET, radiographs (41 patients)
- Malaria Atlas Project: Plasmodium falciparum Parasite Rate (PfPR2-10), 5 km annual maps
- IDC NSCLC Radiogenomics: Non-small Cell Lung Cancer CT, PET (211 patients)
- IDC QIN-PROSTATE-Repeatability: Prostate Cancer MRI (15 patients)