SmartVision lite is a free product trial version primarily for invoice extraction. The launch would follow a multi-phase approach – first would be a soft launch with internal customers primarily the customers who are in active discussions with sales team and internal UST accounts. Within 6 – 8 weeks we will do the final launch from marketing website making it available for global users. This should provide a firsthand product experience for our prospective customers, enable us to keep up with the market trends and also create a competitive advantage.
In 2.6, a major revamp is done on the UI & API architecture end to enable us to scale to the document centric business use cases of our customers.
User will have a provision to define soft validations in addition to hard validations available today. These are warning messages and can be overridden by the user. This is applicable to internal & external validation messages.
Introducing the global model for invoices which supports an expanded list of FOIs & table fields.
In DU, formatter section in Extraction FOI page is enhanced to support 3 types of formatters – Page, Section and Table. User will have the provision to download the available formatters and upload formatters as needed. This allows project specific customizations to meet customer needs.
he accuracy calculation approach for table will be revised to consider the complexity of tables. In 2.6, we will focus on improving the accuracy for simple tables and medium & high complex tables will be targeted in the next releases.
User will have the provision to specify the type of files uploaded to DUT. This could be for Training or Testing. Training files will be used for the model training and testing files will be used to evaluate the model. Default type would be training. Testing files also should be annotated similar to training to capture the ground truth. If testing files are not available, evaluation process would not be triggered and model card would not be available.
Model evaluation would be triggered automatically after training process using the testing files and annotations. Systematic comparison against previous versions will happen and optimal model will be baselined. Only one model version will be retained per project. Model card will be displayed with the latest statistics. It will include the overall model accuracy and the field level accuracy.
DU trainer will have the provision to export files & annotation from one DU trainer project and import them to another DU trainer project. This will allow reuse of files & annotations across DUT projects in same/different environments.
The imported files & annotations will fall into the Annotated tab by default. For an existing DUT project which has annotated files, the import feature will append all the new files & annotation to the existing list. User will be able to modify the annotations on the imported records and save. While importing, user will also have the provision to specify tag for testing/training purpose.