Welcome to Nanonets!
Who is this guide for?
Try Your First Model
When you first open up Nanonets, you'll see a bunch of colorful icons. The icons are different AI models that you can train or use right away. Here are the three ways:
1. Build your own OCR Model
2. Build your Document Classification Model
3. Use a pre-made Nanonets Model
Building Your Own OCR Model
Let's take a fun fictitious example - Say you are starting a new restaurant business, and want to capture all monthly expenses in a CSV at the end of the month.
1. Upload some pages of your document type to teach the AI Model on the Upload Files screen.
We require a minimum of 10 images to train a custom model.
2. Specify the fields that you want the AI to capture from your document on the Specify Fields screen.
In this case, you're labels will look something like this -
Once you're ready, click on Start Training.
3. Show the AI Model where the fields are present on your pages.
On the first page, create rectangular boxes around the text associated with each label. Enter the name of the label (or field) you want to associate the text with, and hit Save. This process is called annotating.
We need a minimum of 10 annotations per label to train a model. We recommend starting with 50 and adding more files depending on the accuracy you see.
It takes about 40 minutes before you can see some magic. Watch your inbox - We'll send you an email once your model is ready!
Best Practices to Train the Model
Refer to an in-depth article here to train the best performance model for you: https://nanonets.com/help/ocr/best-practices-for-high-accuracy-models
Pre-made Nanonets Model
You can quickly use a pre-made Nanonets model for these document types - Invoices, Receipts / Bills, US Driver's Licenses, Passports, Menu Cards. Simply upload your files on the Extract Data screen and review extracted data.
Each of these models is already trained on thousands and thousands of pages and work well out of the box for any language! What's more, that we can quickly enable these models for any additional language.
We frequently add new models to our Product Universe, if you can't find a model for your document type - drop an email to firstname.lastname@example.org and we'll see if we can quickly build one for you.