Presentation: Analyzing images in real-time to extract fields on mobile devices using ultra-fast machine learning
Dec 08, 2022
Next-Gen Applications Stage
- Extracting key-fields from a variety of images and document types remains a challenging machine learning problem.
- The speed-limit for extracting information using cloud services such as AWS and Google Cloud remains in the ~3-5+ seconds range, too slow for utilizing video and augmented reality.
- We propose an ultra-fast on-device solution that extracts fields from invoices and receipts in the sub-second range, with speeds approaching 200ms on modern devices.
- Real-time scanning and extraction open new products and workflows.