Skip to content
PoopCheck PoopCheck
FOR BUSINESS

Annotated stool image dataset

185,000+ real-world stool photos, with AI labels for Bristol Stool Scale type, colour, blood and mucus. Photos from users who consented to research use are available for your ML team or research group.

Request dataset access

What's included

One of the largest consumer stool image datasets in existence, delivered with labels ready for training.

185,000+ labeled images

Real-world stool photos collected through the PoopCheck app, captured across devices, lighting, and demographics.

Bristol Stool Scale class

Every image labeled Type 1–7 on the medically recognized Bristol Stool Scale.

Colour, blood & mucus labels

AI labels with confidence scores for colour, mucus, blood, consistency and shape.

JSON labels + images

Standard delivery format, with custom schemas available. Paired with per-image metadata when applicable.

How our labels are made

AI labels at scale. People check the work.

Three AI models label each photo. Our team then reviews photos, removes the unusable ones and corrects labels, and those corrections go back into training.

  1. PoopCheck app

    People photograph their stool at home and add daily context.

  2. AI labeling

    Three models label Bristol type, color, blood, mucus and more, each with a confidence score.

  3. Human review Non-medical

    Our team removes unusable photos and corrects the labels the models got wrong.

  4. Training data

    Reviewed photos and corrected labels are what our in-house model learns from.

Corrections retrain our model

What teams do with it

Training CV models

Large, diverse, label-ready image set — skip years of collection and annotation.

Academic research

Research-license terms for universities and published studies on gut health and digestive medicine.

Model benchmarking

Benchmark your model against an independent set of real-world photos before you ship it.

Benchmark data

A real-world evaluation set for gut-health AI, not another lab-sanitized collection.

Annotation methodology

Three vision models (our in-house model plus two frontier models) label each photo and vote, and each label ships with a confidence score, so you can decide how strictly to filter for training versus evaluation. Our team then reviews photos, removing unusable ones and correcting labels. Review is non-medical: labels are not clinician-annotated.

Licensing

Research licenses are available for academic use with publication rights. Commercial licenses cover product development, with terms scaled by subset size and use case. We'll send a draft agreement after an intro call.

Frequently asked questions

How is the dataset licensed?
We offer research licenses for academic use (with publication rights) and commercial licenses for product development. Terms vary by use case — contact us and we'll scope a license that fits.
How is the data delivered?
Default delivery is a signed S3 bucket with images and JSON label files, plus a schema document. We can accommodate custom formats (COCO, CSV, per-label folders) on request.
Can we get a custom subset?
Yes. Common requests include "only Types 3–5", "only flagged samples", or balanced gender/age splits. We can scope these from the master dataset.
How is the data de-identified?
Standard deliveries contain only the stool photo and its labels: no names, emails, account IDs or timestamps. Every image is re-encoded to strip camera and location metadata, and files are renamed so nothing identifying leaks through file names.
What are the usage restrictions?
Standard restrictions: no redistribution outside your organization, no resale, and no use that would enable re-identification. Full terms are in the license agreement we send before delivery.
How is it priced?
Pricing depends on subset size, annotation depth, and whether the license is research or commercial. Contact us with the specifics of your need and we'll send a quote.

Also available

Stool analysis API

Send a photo, get Bristol type, colour, consistency, mucus and blood indicators back as JSON.

PoopCheck for Business

Overview of everything we license — API access, annotated data, and how teams build on it.

Request dataset access

Tell us about your use case and any specific subset needs. We'll reply within one business day.

0 / 2000

Or email us directly at marco@softallthings.com.

4.4 · Get the app free