Spot the mask

Spot the mask challenge

GithubOpen In Colab


This is a knownledge based challenge on zindi (opens in a new tab) that aimed to help beginers in Machine Learning and Deep Learning to practise their skills with real world examples.

Challenge Description

The task involves developing a machine learning model capable of accurately identifying whether a person in an image is wearing a mask or not. Participants will have access to a labeled dataset consisting of images with corresponding binary labels (1 for mask-wearing, 0 for no mask).

Evaluation Metric

The competition will be evaluated based on the Area under curve (opens in a new tab). The Area under curve is a metric that considers both recall(TPR) and Fall Out (FPR), making it suitable for binary classification tasks.

True Positive Rate (TPR)

The True Positive Rate (TPR), also known as sensitivity or recall, is a key metric in binary classification. It is defined as:



  • TP is the number of true positives, i.e., the instances correctly predicted as positive.
  • FN is the number of false negatives, i.e., the instances incorrectly predicted as negative.

TPR provides insights into the model's ability to correctly identify positive instances out of all actual positive instances.

In the context of a Receiver Operating Characteristic (ROC) curve, TPR is often plotted against the False Positive Rate (FPR) to evaluate the model's performance across different threshold settings.

False Positive Rate (FPR)

The False Positive Rate (FPR) is a crucial metric in binary classification. It is also known as the fall-out or Type I error rate. FPR is defined as:



  • FP is the number of false positives, i.e., the instances incorrectly predicted as positive.
  • TN is the number of true negatives, i.e., the instances correctly predicted as negative.

FPR provides insights into the model's tendency to produce false alarms or positive predictions when the true class is negative. A low FPR is desirable, as it indicates a lower rate of false positives and, consequently, a better model performance in correctly classifying negative instances.


The dataset provided for this challenge includes a diverse set of images containing people with and without masks. Participants are encouraged to explore and preprocess the data to enhance model performance.

Data Files

  • This archive contains the set images.
  • train_lables: The labels for the image zip.
  • SampleSubmission: Use this for taking the image for submission.

Getting Started

To participate in the challenge, follow these steps:

  1. Sign up for Zindi: If you don't have an account on Zindi Africa, sign up at Zindi (opens in a new tab).

  2. Join the Competition: Navigate to the Spot the Mask Challenge (opens in a new tab) page and click "JOIN."

  3. Download Data: Access the provided data files from the competition page and unzip them locally.

  4. Explore the Data: Understand the structure of the dataset, inspect sample images, and consider potential preprocessing steps.

  5. Build Your Model: Develop a machine learning model capable of predicting whether an individual is wearing a mask based on the provided images.

  6. Make Submissions: Upload your predictions on the test set to the Zindi platform and receive instant feedback on your model's performance.

  7. Collaborate and Learn: Engage with the competition community, discuss approaches, and learn from others to improve your model.