Pumapalya ang basic OCR-to-calendar tools sa real-world flyers at mobile screenshots dahil tina-treat nila ang image bilang flat text at blindly kina-copy ang first line (lines[0]) bilang event name. Sa phone screenshot o concert poster, ang first line ay halos palaging status-bar noise (9:41 AM 5G), ticketing watermark (TICKETMASTER), o promoter intro (LIVE NATION PRESENTS).
Naiintindihan ng 3-Stage Local AI Pipeline ng Utiliome ang visual hierarchy at semantic entities ng flyer mo:
- Tesseract v6 Layout Block at Bounding-Box Extraction: Kina-extract ang exact pixel coordinates ng bawat line (
bbox: { x0, y0, x1, y1 }) at kina-calculate ang visual font height (y1 - y0) relative sa median text height ng poster. - Local
Xenova/bert-base-NERWeb Worker: Ni-normalize ang all-caps poster text sa smart title case at nagpapatakbo ng token classification sa loob ng Web Worker para i-score angLOC(venues, streets, cities),ORG(bands, hosts),PER(artists, speakers), atMISC(tours, festivals) sa loob ng 90ms. - Headline at Venue Pairing: Pina-pair ang adjacent high-font-ratio headline lines sa malinis na
Event Titlehabang mina-merge ang consecutiveLOClines (tulad ng venue name sa itaas ng street address) saLocation / Venuefield.