Open RafaPolit opened 2 years ago
Additional context as to why the click-to-fill matters: since we use the click-to-fill rectangles as an input to the information extraction service, this could cause the machine learning algorithm to get wrong training examples.
Describe the bug Triple clicking a line in the PDF visually selects only one row (a title for example), but upon creating a text reference, the selection rectangles sent actually encompass the entire page.
To Reproduce Steps to reproduce the behavior:
Expected behavior The sent rectangles should match the selected area.
This affects the "click to fill" flow as well
Additional context Both of the PDFs from the E2E puppeteer fixtures display this behavior in case this is somehow PDF dependent. This may just be a bug with our boundary rectangle detecting library, so this may be out of our control. If so, please create a bug report with the developers of the library.