[Approach 1]

if: CNN classifier

1. get the bbox.
2. crop the bbox with OpenCV.
3. run OCR for that region.

else:

1. Get the important keywords coordinates.
2. Run OCR for the increased (let say 20% of width) bbox area to get the numerical data.

RUN

python train.py --img 1200 --batch 4 --epochs 500 --data train/data.yaml --cfg models/yolov5l.yaml --weights models/yolov5l.pt