Computer Vision Mini Lab
One image. Three questions. Explore classification, object detection, and segmentation directly in your browser.
What are we asking the computer?
Use the same image for all three experiments.
- Classification: What is in the image?
- Object detection: What is there, and where?
- Segmentation: Which pixels belong to the objects?
1. Choose an image
No image selected. Choose an image to begin.
2. Ask the computer
Your image will appear here.
3. Results
Classification
A label for the whole image.
Not run yet.
Object Detection
Classes plus bounding boxes.
Not run yet.
Segmentation
Which pixels belong to each object? The masks are shown over the image.
Not run yet.
What changed?
The image did not change. The question changed.
Classification → WHAT
Detection → WHAT + WHERE
Segmentation → WHICH PIXELS
More spatially precise questions require more spatially precise outputs.
Technologies used
This demonstration uses Hugging Face Transformers.js to run pretrained machine-learning models in the browser.
- Transformers.js — browser-based machine learning.
- ONNX Runtime — inference runtime used by Transformers.js.
- Hugging Face Hub — source of the pretrained model files.
- HTML, CSS and JavaScript — static web interface.
Important notes
⚠️ This may load your browser. Running machine-learning models locally can use substantial CPU, GPU, memory and battery resources. Larger models may take time to initialise.
⏳ The first run may be slow. A model is downloaded when a task is first used and may be cached by the browser afterwards.
🌐 Internet connection required. This page loads the JavaScript library and pretrained models from external services when they are not cached.
🔬 Educational demonstration. These are general-purpose pretrained models. Predictions can be wrong; confidence scores are not guarantees.
📷 Privacy. Avoid sensitive or personally identifiable photographs in classroom demonstrations. The inference is performed in the browser, while model files are downloaded from Hugging Face.