🌈 🍋 🍈 🥒 🍏 🍎 🫑 🍅 🌶️ 🍊 🍐 🍇 🍋‍🟩 🌈

The Living Optics Hyperspectral-Fruit dataset is now available to download along with our new spatial-spectral machine learning project .

Follow Living Optics on Hugging Face for the latest developments in Computer Vision, Machine Learning and Hyperspectral Imaging.

The dataset is the first of its kind combining RGB and hyperspectral data with instance segmentation masks and detailed subclass labels.

What is hyperspectral data?

Hyperspectral data provides a detailed spectral signature for each sampled pixel in the scene.
The Living Optics Camera provides 96 channels of colour information spanning from blue through to near-infrared a.k.a. VIS-NIR light.

Spectral samples are evenly spaced across the scene in a sparse grid, allowing us to measure which frequencies of light are being reflected and absorbed by the different materials in the scene.
This allows you to distinguish between different varieties of fruits, without relying on spatial information.

What's in the dataset?

100 hyperspectral images along with instance segmentation masks for 19 different object classes. Check out the dataset card for the full details.

Among the classes there are:

• 3 varieties of peppers 🫑🌶️
• 5 varieties of apples (including a fake, plastic apple) 🍏🍎
• 3 varieties of tomatoes (including a fake plastic tomato) 🍅
• 2 varieties of oranges 🍊

This allows you to explore how spectral information can be used to produce more accurate subclass classifications.