Hi,
Thanks for the great work on this repo. It has saved a lot of my time. I'm planning to use your train_ocr.py script to see if I can either develop a OCR model from scratch or fine tune on my target domain.
1) For fine-tuning - where can I find this base model mentioned textspotter/tmp/DS_CVPR/backup2/ModelResNetSep2_25000.h5
2) For training from scratch -
a) is it better to train region detection + ocr in parallel like your train.py script or separately train an OCR model and use it as starting point for train.py. any thoughts
b) If latter, can I use the output model of `train_ocr.py` as an input to `train.py` 's `--model` param ?
Hi, Thanks for the great work on this repo. It has saved a lot of my time. I'm planning to use your
train_ocr.pyscript to see if I can either develop a OCR model from scratch or fine tune on my target domain.1) For fine-tuning - where can I find this base model mentioned
textspotter/tmp/DS_CVPR/backup2/ModelResNetSep2_25000.h52) For training from scratch - a) is it better to train region detection + ocr in parallel like your
train.pyscript or separately train an OCR model and use it as starting point for train.py. any thoughtsAny help appreciated, Thanks.