Skip to main content

Collect Data

Data is the foundation of any machine learning model. Before you can train a model, you need a labelled dataset that represents the different classes you want to recognise.

During this bootcamp we collect image data — either by uploading existing images, taking photos directly from your laptop camera, or using your mobile phone by scanning a QR code. No CLI or terminal is needed.

What You're Collecting​

Each sample is an image paired with a label — the class name that describes what's in it.

For example, if you're building an image classifier, you might collect samples labelled cat, dog, and background.

Methods​

1. Upload Images from Your Computer​

If you already have a folder of images ready:

  1. In Edge Impulse Studio, go to Data acquisition.
  2. Click Upload data.
  3. Select your image files (JPG, PNG).
  4. Set the label for the batch.
  5. Choose whether to add them to the Training or Test set.
  6. Click Upload — your images will appear in the dataset immediately.

2. Take Photos Using Your Laptop Camera​

You can capture images live directly from your laptop's webcam without leaving the browser.

  1. In Edge Impulse Studio, go to Data acquisition.
  2. Click Collect data and select your laptop as the device.
  3. Set the label for what you're about to photograph.
  4. Point your laptop camera at the subject and click Start sampling.
  5. Edge Impulse will capture frames and add them to your dataset.

3. Use Your Mobile Phone via QR Code​

Edge Impulse lets you use your phone as a camera to capture images directly into your project.

  1. In Edge Impulse Studio, go to Data acquisition.
  2. Click Connect a device → Use your mobile phone.
  3. A QR code will appear on screen — scan it with your phone's camera.
  4. Your phone opens Edge Impulse in the browser and connects to your project automatically.
  5. Set the label, point your phone camera at the subject, and tap Start sampling.
  6. Photos are sent directly to your Edge Impulse dataset in real time.
tip

The mobile phone method is great for capturing images from different angles and distances quickly — your phone camera is also higher quality than most laptop cameras.

Splitting Data: Training vs Test​

Edge Impulse splits your dataset into two sets:

  • Training set — used to teach the model.
  • Test set — used to evaluate the model after training. Keep this separate and don't train on it.

A good starting split is 80% training / 20% test. Edge Impulse handles this automatically when you upload data.

Tips for Good Image Data​

  • Photograph your subject from multiple angles — front, side, top.
  • Vary the lighting and background across samples.
  • Make sure the label matches exactly what's in the image — no mislabelled samples.
  • Aim for at least 30–50 images per class to start; more is better.
  • Avoid using the same photo repeatedly — variety is what makes a model robust.

Your Turn​

  • Open Data acquisition in Edge Impulse Studio.
  • Collect at least 3 classes of images using any of the methods above.
  • Aim for a minimum of 20 images per class.
  • Verify your images appear in the dataset with the correct labels.
  • Check that your dataset is split into Training and Test sets.