Comprehensive testing video
#8
by ritheshSree - opened
Congratulations on a powerful OCR Model
I did some comprehensive testing and here is the video: https://youtu.be/nHIaLWHtcag
I did find some issues with handwritten data
Hi,
Thank you for the thorough testing!
I would like to point out that the current version of the model is primarily trained on PDF data, which means handwritten data and screenshots are out of distribution. However, the model still performs reasonably well on these out-of-distribution tasks. We have also shared a finetuning notebook where the model is finetuned on handwritten data. You can try it out here.