---
title: "Image Data Collection for Hair Loss Classification Task"
description: "With clear guidelines and a sharp execution strategy, we delivered a high-quality dataset tailored for hair loss classification tasks."
url: "https://unidata.pro/cases/image-data-collection-for-hair-loss-classification-task/"
date_modified: "2026-02-11T11:11:08+03:00"
language: "en-US"
---
The Task
--------

Imagine this: A medical company conducting advanced research on baldness approaches us with an unusual request.

They needed a dataset of bald men. Each set had to consist of five photos of the same person taken from different angles: front view, profile (both sides), top view, and back view.

But it wasn’t just about collecting photos—it was about creating a database where each image was "labeled" with medical precision. The reference point was the **Norwood scale**, a classification system for the stages of baldness. There was no room for error.

The Solution
------------

### Preparation Stage:

- **Studying the Norwood Scale:** We dived deep into the topic, researching everything available about the scale.
- **Guideline Creation:** With the help of medical experts, we developed an extremely detailed guide for our annotators, describing each stage of baldness down to the millimeter, complete with illustrations and diagrams.

### Annotator Training and Preparation:

- Every annotator had to pass an entry test on the Norwood scale (a 50-example test) before starting work.
- We conducted hands-on training on annotating photos with different degrees of baldness, analyzing ambiguous cases, and discussing evaluation criteria.
- After training, we calibrated the annotators’ work to ensure uniformity and precision.
- Annotators were also trained in using spreadsheets, data processing platforms, and annotation tools.
- To support the team, we set up a helpdesk for real-time expert assistance.

### Data Collection:

- A major crowd-platform served as the primary channel for collecting photos.
- We created a clear and engaging task for participants, complete with detailed instructions and examples. To encourage high-quality submissions, we offered competitive compensation.
- Our team closely monitored the process to promptly identify and address potential issues.

### Data Validation:

- We implemented a multi-layered verification system. First, an automated check ensured that all required photos were present and met technical specifications.
- Then, experts manually annotated the images using the Norwood scale, cross-referencing them with our guidelines.
- To guarantee accuracy, we conducted a **cross-validation**, where one set of experts reviewed the work of others.

The Result
----------

- Not only did we meet the deadline, but we also achieved **near-perfect annotation accuracy**. Thanks to thorough annotator training and a multi-step validation process, annotation errors were kept to a minimum—**less than 1%**.
- The client approved all the data and was highly satisfied with the results.





[View as Markdown](https://unidata.pro/cases/image-data-collection-for-hair-loss-classification-task.md)

## Hero

**Industry and use case:** Medicine **Data:** 200 annotated sets **Project duration:** 1.5 months

## Main title

Image Data Collection for Hair Loss Classification Task

## Description

We found a fast and efficient solution for gathering 200 sets of photos of bald men from different angles. Thanks to a detailed guideline, we annotated the images according to the Norwood scale with the highest accuracy.