smallNORB is a dataset for recognizing 3D objects based on their shapes. The dataset contains images of 50 toys belonging to 5 general categories: four-legged animals, human figures, airplanes, trucks, and cars.

The objects were captured by two cameras under 6 lighting conditions, 9 elevation angles (every 5 degrees from 30 to 70 degrees), and 18 azimuth angles (every 20 degrees from 0 to 340 degrees). The training set consists of 5 instances in each category (instances 4, 6, 7, 8, and 9), and the test set consists of the remaining 5 instances (instances 0, 1, 2, 3, and 5).