Task: solve OCR problem for captchas.
My solution is partially based on paper.
Dataset is available here.
All necessary scripts are in directory utils
:
utils/dataset.py
- file with dataset;utils/model.py
- file with OCRmodel;utils/utils.py
- file with useful utils;utils/train.py
- file with training script;
Experiments were coducted in ocr solution.ipynb
, so find results there. Final CER on test data was around 5.3%.