Dataset Card for Food Waste Dataset

This is a FiftyOne dataset with 375 samples focused on food waste analysis and nutritional content detection.

Installation

If you haven't already, install FiftyOne:

pip install -U fiftyone

Usage

import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("andandandand/food-waste-dataset")

# Launch the App
session = fo.launch_app(dataset)

Dataset Details

Dataset Description

This dataset contains detailed information about food waste, combining visual data with comprehensive nutritional measurements. Each sample includes an image of a meal along with ingredient-level nutritional information measured both before and after consumption, enabling food waste analysis and nutritional content detection.

The dataset has been enhanced with:

YOLO-E segmentation for ingredient detection and segmentation
DINOv2 embeddings for visual similarity analysis
Translated ingredient names from German to English
Nutritional metadata including calories, fats, proteins, carbohydrates, and salt content
Curated by: L. Stroetmann, a la QUARTO, AI Service Center at HPI (Hasso Plattner Institute), Voxel51
Enhanced by: FiftyOne computer vision pipeline
Language(s): English (translated from German)
License: MIT

Dataset Sources

Original Repository: AI-ServicesBB/food-waste-dataset
Processing Code: Available in the accompanying Jupyter notebook
Enhanced Version: Includes segmentation masks and embeddings

Uses

Direct Use

This dataset is suitable for:

Food waste analysis and sustainability research
Nutritional content detection from images
Ingredient segmentation and recognition
Computer vision model training for food-related tasks
Multi-modal learning combining visual and nutritional data
Food portion estimation and consumption analysis

Out-of-Scope Use

This dataset should not be used for:

Medical diagnosis or personalized dietary recommendations
Commercial food recognition without proper validation
Applications requiring real-time nutritional analysis without expert oversight
Any use that could promote harmful eating behaviors

Dataset Structure

The dataset contains 375 samples split into train and test sets, with each sample containing:

Image Data

filepath: Path to the meal image
metadata: Image dimensions, format, and technical details

Nutritional Information (Per Ingredient)

ingredient_name: Name of each ingredient (translated to English)
article_number: Unique identifier for ingredients
number_of_portions: Portion count
weight_per_portion: Weight per individual portion
weight_per_plate: Total weight on plate
kcal_per_plate, kj_per_plate: Caloric content
fat_per_plate, saturated_fat_per_plate: Fat content
carbohydrates_per_plate, sugar_per_plate: Carbohydrate content
protein_per_plate: Protein content
salt_per_plate: Salt content

Before/After Consumption Measurements

weight_before/after: Total meal weight
kcal_before/after: Total calories
fat_before/after: Total fat content
carbohydrates_before/after: Total carbohydrates
protein_before/after: Total protein
salt_before/after: Total salt

Food Waste Metrics

return_quantity: Amount of food returned/wasted
return_percentage: Percentage of food wasted

Computer Vision Annotations

yoloe_segmentation: Ingredient segmentation masks from YOLO-E
segment_embeddings: DINOv2 embeddings for segmented regions
dinov2-image-embeddings: Full image embeddings
similarity indices: For content-based search and analysis

Dataset Creation

The Google Colab notebook used to curate and produce the dataset is available here:

Curation Rationale

This dataset was created to support research in food waste reduction and nutritional analysis. By combining visual data with detailed nutritional measurements, it enables the development of computer vision systems that can:

Automatically detect and quantify food waste
Estimate nutritional content from images
Analyze consumption patterns
Support sustainability initiatives in food service

Source Data

https://huggingface.co/datasets/AI-ServicesBB/food-waste-dataset

Data Collection and Processing

The original dataset was collected by the L. Stroetmann, a la QUARTO, and the AI Service Center at HPI and contained:

Images of meals in German food service settings
Detailed nutritional information in German
Before and after consumption measurements

Processing steps included: 3. Embeddings: DINOv2 model used for visual feature extraction 4. Similarity indexing: Computed for both full images and segmented regions

Translation: German ingredient names and field names translated to English
Segmentation: YOLO-E model applied for ingredient detection
Metadata computation: Image technical details extracted

Who are the source data producers?

The original data was produced by the AI Service Center at the Hasso Plattner Institute (HPI) as part of food waste research initiatives.

Annotations

Annotation process

Ingredient Translation: Manual mapping of 40+ German ingredient names to English equivalents
Segmentation: Automated using YOLO-E model trained on food ingredients
Embedding Generation: Automated using DINOv2 vision transformer
Quality Control: Visual inspection of segmentation results

Who are the annotators?

Translation: Manual annotation by dataset curator
Segmentation: YOLO-E model (yoloe-11s-seg.pt)
Embeddings: DINOv2-ViT-L14 model

Technical Details

Ingredients Covered

The dataset includes 40+ food ingredients including:

Proteins: meatballs, fish fillet, chicken, beef, pork, sausages
Carbohydrates: rice, potatoes, bread dumplings, spaetzle
Vegetables: green beans, carrots, cabbage, cauliflower, peas
Sauces and condiments: various gravies, mustard sauce, dressings
Dairy: cream, vegetable-based cream alternatives

Model Performance

The dataset includes pre-computed:

Segmentation masks with ingredient-level precision
Visual embeddings enabling similarity search
UMAP visualization for dataset exploration

Bias, Risks, and Limitations

Limitations

Cultural bias: Dataset reflects German food service context
Ingredient coverage: Limited to ~40 common ingredients
Portion size: Focused on institutional serving sizes
Image quality: Consistent lighting/background conditions
Temporal scope: Snapshot data, not longitudinal study

Risks

Nutritional accuracy: Automated estimates should not replace professional dietary advice
Generalization: Model performance may vary on different food cultures/preparations
Privacy: While anonymized, institutional food service data patterns might be identifiable

Recommendations

Users should:

Validate nutritional estimates with professional dietary knowledge
Consider cultural context, this dataset was collected in Germany
Use appropriate evaluation metrics for food waste applications
Acknowledge dataset limitations in publications and applications

Citation

If you use this dataset, please cite both the original source and the enhanced version:

Original Dataset:

@dataset{hpi_food_waste_2024,
  title={Food Waste Dataset},
  author={Felix Boelter and Felix Venner},
  year={2024},
  url={https://huggingface.co/datasets/AI-ServicesBB/food-waste-dataset}
}

Enhanced Version:

@dataset{food_waste_fiftyone_2024,
  title={Food Waste Dataset with FiftyOne Enhancements},
  author={Antonio Rueda-Toicen},
  year={2024},
  url={https://huggingface.co/datasets/andandandand/food-waste-dataset}
}

More Information

For technical details about the processing pipeline, see the accompanying Google Colab notebook. The dataset supports various computer vision tasks and can be explored interactively using the FiftyOne application.

Related Work

FiftyOne: Open-source tool for dataset curation and model analysis
YOLO-E: State-of-the-art object detection and segmentation
DINOv2: Self-supervised vision transformer for embeddings
Food waste reduction and sustainability research

Dataset Card Contact

Antonio Rueda-Toicen

For questions about the original dataset, please refer to the AI Service Center, HPI.

image imagewidth (px) 1.69k 1.69k