Thank you for your help on our annotation task!

Contents

1. Using the tool
2. Annotation guidelines

Using the tool

You will be presented with an image (centre), a set of available field labels (right), and an OCR results pane (left).

Select a target field type (such as email, address, etc) on the right, then draw bounding boxes over the image to indicate where the data is visible in the page.

To highlight non-rectangular regions, draw multiple, overlapping bounding boxes of the same label class. These will be consolidated into one field automatically in the OCR results pane.

(If the tool has loaded correctly) a new item will appear in the OCR results pane for each region you annotate on the image - including the OCR extracted words at that location, and radio buttons for you to indicate whether the transcription was:

• "Correct"
• "Wrong" (and you're able to confidently provide a correct transcription), or
• "Unclear" (you're not able to clearly read what the result should be - due to poor handwriting or similar)

If you select "Wrong", you'll also be prompted to enter the correct transcription for that section. When you've annotated all fields for the page and reviewed all OCR results, go ahead and click "Submit" (bottom right) to move on to the next task.

Although the framework for this tool is a stable service, the specific OCR workflow here is a custom prototype and therefore may still have some bugs. If you encounter problems with, for example, the task not loading correctly or the tool not working as described, please try:

• Refreshing the page
• Exiting/returning the task (not submitting) and then "start working" from the task list again