Neural End-to-end Kuzushiji OCR
The image size is 576 x 576 All characters detected is in charlist.csv (409 characters at the moment)
I am thinking about adding config.py to put all variables there.
When doing OCR, just put images in decoder/source and run ocr.py in ocr.py there are 3 steps
- source image resize to 576
- detection with torch
- clustering with dbscan
all output are in decoder/ocr/