MobileMold: A Smartphone-Based Microscopy Dataset for Food Mold Detection

Dinh Nam Pham¹, Leonard Prokisch², Bennet Meyer³, Jonas Thumbs⁴

¹ Technical University of Berlin, ² University of Regensburg, ³ ETH Zurich, ⁴ University of Tübingen

           

This is the oficial repository of MobileMold, a smartphone-microsope-based dataset with 4941 annotated images for food mold detection.

🌟 About MobileMold

MobileMold is a comprehensive dataset comprising 4,941 annotated images for food mold detection, captured using smartphones with various clip-on microscope attachments. The dataset addresses the growing need for accessible, low-cost food safety monitoring by leveraging smartphone-based microscopy. This enables research and development in computer vision applications for mold detection on various food surfaces.

---

📊 Dataset Overview

• Total Images: 4,941
• Annotations: Food Type and Mold Label
• Food Types: 11 categories (carrot, orange, creamcheese, tomato, toast, raspberry, mixed bread, blackberry, blueberry, cheese, onion)
• Microscope Types: 3 different clip-on smartphone microscopes (30x-100x magnification)
• Smartphones: Images captured with 3 different smartphone models

---

📢 Data Release

You can download the full dataset here:

• Kaggle
• Hugging Face
• Zenodo

---

📁 Dataset Structure

MobileMold/
├── metadata.csv # Complete dataset metadata (4,941 entries)
├── train_metadata.csv # Training split metadata
├── val_metadata.csv # Validation split metadata
├── test_metadata.csv # Test split metadata
├── original/ # Original microscope images (as captured)
│ ├── L10 - 48.jpeg
│ ├── L10 - 25.jpeg
│ ├── L10 - 161.jpeg
│ └── ... (4,941 files total)
└── cropped_resized/ # Preprocessed images (same filenames)
├── L10 - 48.jpeg # Cropped to mold region & resized
├── L10 - 25.jpeg
├── L10 - 161.jpeg
└── ... (4,941 files, 1:1 mapping to original/)

---

📊 Dataset Composition

Image Versions

1. original/ - Raw images as captured by smartphone microscopes

   • Various resolutions (depending on smartphone and microscope)
   • Full field-of-view including background
   • Unprocessed image data

2. cropped_resized/ - Processed images

   • Cropped to focus on mold regions
   • Resized to consistent dimensions
   • Same filenames as original folder

Metadata Format

Each CSV file contains the following columns:

Column	Description	Values/Examples
filename	Image filename (same in both folders)	L10 - 48.jpeg
mold	Binary indicator of mold presence	True / False
food	Type of food in image	carrot, bread, cheese, tomato, etc.
phone	Smartphone model used	iPhone SE 2nd Generation, etc.
microscope	Clip-on microscope model	Apexel 100x, etc.

Example metadata entry:

filename,mold,food,phone,microscope
L10 - 48.jpeg,True,carrot,iPhone SE 2nd Generation,Apexel 100x

FreshLens Mobile App

The freshlens-app repository contains a Flutter-based mobile app designed for consumer-facing demonstrations and can be used in conjunction with a hosted model. Using a smartphone microscope attachment, users can capture or import images of food. The app then displays the probability that the food is moldy.

Citation

If you use this useful for your research, please cite this as:

@inproceedings{Pham2026MobileMold,
author = {Pham, Dinh Nam and Prokisch, Leonard and Meyer, Bennet and Thumbs, Jonas},
title = {MobileMold: A Smartphone-Based Microscopy Dataset for Food Mold Detection},
year = {2026},
isbn = {9798400724817},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3793853.3799806},
doi = {10.1145/3793853.3799806},
booktitle = {Proceedings of the ACM Multimedia Systems Conference 2026},
pages = {402–408},
numpages = {7},
keywords = {Dataset, Smartphone, Food, Mold, Microscope, Mobile, Fungal},
series = {MMSys '26}
}

📄 License

This dataset is available under the terms of the CC BY-NC 4.0