evidoria

← Back to browse

Good practice Imported

Seeing AI — Microsoft's Real-Time AI Visual Assistant for Blind and Low-Vision Users

United States of America · Redmond · See the United States of America profile · See the Redmond profile

Evidence: Descriptive / self-reported Top 10% 96/100 · Ask Evidence Copilot about this practice

Microsoft's free Seeing AI app narrates text, currency, people, barcodes and scenes in real time through a phone camera for blind and low-vision users; it reached 35+ countries and completed over 3 million tasks within its first six months.

3,000,000+ tasks
Tasks completed (first six months post-launch (2017))
35+ countries
Countries reached (2017)
19 languages
Languages supported (current) (current)
4.3 out of 5
App Store rating (current)
3.57 % error rate
Image classification top-5 error rate (Microsoft research benchmark)
13,000+ entries
Hackathon entries surpassed (2015 internal hackathon)
Seeing AI — Microsoft's Real-Time AI Visual Assistant for Blind and Low-Vision Users

Details

Maturity
Established
Promoter
Microsoft
Period
2017–present
Keywords
assistive technology, artificial intelligence, mobile apps, accessibility

Context

Seeing AI is a free iOS app from Microsoft Research and the Microsoft Garage, grown out of a 2015 internal hackathon that beat more than 13,000 other entries, released broadly on 12 July 2017.

Objectives

The app aims to turn a phone camera into a talking assistant for blind and low-vision users, narrating text, currency, people, barcodes and scenes in real time.

Activities

It combines on-device and cloud computer vision to read text aloud with alignment guidance, identify currency and barcodes, recognise saved contacts, describe facial expressions and produce spoken scene descriptions.

Results

By the end of 2017 Microsoft reported over 3 million completed tasks and availability in 35+ countries; it won the American Foundation for the Blind's Helen Keller Achievement Award in 2018. The app now supports 19 languages and holds a 4.3/5 Apple App Store rating; Microsoft's underlying image-classification research reported a 3.57% top-5 error rate.

Conclusions

The 3-million-task and 35-country figures are Microsoft's own self-reported usage statistics rather than independently audited numbers, and the accuracy figures describe general research benchmarks rather than a controlled study of real-world task success for blind users.

Implementation
Indicative cost
High (€500k–€5M)
Time to results
Long (> 3 years)

Commonly funded by

Own resources / municipal budget

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful