Towards Accurate Scene Text Recognition with Semantic Reasoning Networks

This is an unofficial implementation of SRN model for Pytorch.The model has reached the results in the paper on some datasets, and some datasets have not yet reached. Instead of training PVAM and other modules in stages as described in the paper, we train the model end-to-end. The whole project refers to Paddle-OCR.

Results

IIIT5K	ic13_1015	ic03_867	ic15_1811	svt_647	svtp_645	cute80_288
94.6	92.4	94.0	80.0	91.7	83.4	84.0

Requirements

Pytorch >=1.2.0

Training

bash train.sh 

--train_data_dir is the train dataset, download the train dataset from ASTER

Eval

bash test.sh

--test_data_dir is the eval dataset path.

Reference

PaddleOCR
ASTER