SmartVision Release 2.7

Overview

In 2.7 release, we are targeting to launch SV Online from UST marketing website. This would allow global business users to explore SmartVision invoice extraction capabilities.

Core capability enhancements in DU include – extraction from Barcode/QR Code, extraction from Logos/Image.

DU Trainer is enhanced to include assisted continual learning capabilities which should significantly reduce the annotation time. Users will also have the provision to choose the OCR engine for training.

In 2.7 release, SmartVision online will be focusing on the maintenance, admin related activities for the final stage.

Following features will be covered in 2.7:

• Report a Problem
• Feedback form
• Support for tiff files
• Features to support maintenance activities – reports, delete expired accounts & data
• PII encryption
• Multi-browser support – Firefox

Capability to extract data from barcode/QR codes in documents. In 2.7, scope is limited for invoice fields corresponding to B2B E-Invoice in India. Pre-built post processors would map the extracted fields to the relevant output fields. These fields will be displayed in Insights section in document review screen.

• GSTIN of supplier
• GSTIN of Recipient
• Invoice number as given by Supplier
• Date of generation of invoice
• Invoice value (taxable value and gross tax)
• Number of line items
• HSN Code of main item (the line item having highest taxable value)
• Unique Invoice Reference Number (IRN)

Ability to process multiple barcodes/QR codes & tweak post processors to meet specific customer requirements is planned in next release.

Capability to extract vendor names from logos & images. We will be leveraging Azure API services to perform this extraction. This is expected to improve the extraction accuracy of Vendor name extraction.

This release includes 3 customer specific invoice models with improved accuracy. These models were re-trained for additional fields & new templates.

Changes to enable processing of unstructured documents (Contracts) with more than 25 pages.

Adding the capability to automatically annotate the files using the base model selected in DU Trainer. After uploading the files for annotation, users can trigger “Run prediction”. This would automatically annotate the files and move them to the annotated tab. User will have the provision to further review & correct the annotations if needed.

This would significantly reduce the annotation time and also provide opportunities to fine tune the base model.

Consider a scenario, were an existing model with 15 fields needs to be extended for 2 additional fields. Users can

1. Create a new DU trainer project with the existing model as the base model
2. Upload the files for annotation
3. Run prediction - this would automatically annotate all the files for all the 15 fields already available in the model and move them to annotated tab.
4. Review the annotations from annotated tab, make corrections if needed & add the 2 new fields
5. Run model training - A new model with all the 17 fields will be created.
6. Refer model cards for accuracy statistics

User will have the provision to choose OCR engine while creating the training project, model training will be based on the selected engine.

• (8323) IQVIA : DU extraction NoSuchKey error

Release Artifacts

Deployment Artifacts

For Support and more enquiries you can write a mail to us on feedback.smartops@ust-global.com