Hey everyone,
I’ve been exploring different AI-based Optical Character Recognition (OCR) techniques to extract and process structured data from scanned utility bills. Many utility companies still provide bills in PDF or image formats, which makes automation challenging.
I was wondering if anyone here has experience implementing NVIDIA’s AI-powered OCR models for similar tasks? Specifically:
- What are the best models for recognizing complex text structures on utility bills?
- How effective is TensorRT for optimizing OCR inference speed?
- Any experience with NVIDIA’s TAO Toolkit for training custom OCR models?
I recently worked on a project related to utility bill checking for FESCO in Pakistan fescobills.net.pk, and I see a huge potential in automating data extraction from these bills. Would love to hear your thoughts on the best practices!