ORBIT Dataset

A disability-first machine learning dataset for teachable object recognition, contributed by people who are blind or have low vision. The original ORBIT dataset (Massiceti et al., 2021) contains 3,822 videos of 486 objects from 67 data collectors, predominantly in the UK and Canada. A follow-on ORBIT-India dataset (India et al., 2026) extends the protocol to Indian contexts with 105,243 images of 76 objects from 12 blind or low-vision collectors. ORBIT is notable for centring real-world conditions experienced by blind users — poor framing, occlusion, motion blur, low light — rather than the cleaner imagery typical of sighted-collected datasets.