Closed mwharton3 closed 4 years ago
Rotation, document vs photo, and stretching are complete (definitely room for improvement on the model though). Framework for the others is in place, but not implemented (need to operate on patches, not rescaled images).
Need a very basic but well-trained/selected convnet to detect issues and apply labels to image files. e.g.:
In order to keep fast, thinking of operating on thumbnails and patches from the original resolution.