This PR addresses issue #2 .
I have preprocessed the images to enable training only on the form and main features not the extra spaces around the form. Images have been resized to (430, 250) to retain main features.
This with a CNN-based model gives 95.6% accuracy with no over-fitting.
This PR addresses issue #2 . I have preprocessed the images to enable training only on the form and main features not the extra spaces around the form. Images have been resized to (430, 250) to retain main features. This with a CNN-based model gives 95.6% accuracy with no over-fitting.
Please review this PR.