A CRNN python implement based on TensorFlow
- TensorFlow r1.0
- lmdb lib
- OpenCV2
Code from Chaoyue Wang
03/24/2017 Update:
1.Added utility.py to handle check points and string conversion.
2.Added check point system to keep model.
02/25/2017 Update:
1.Tested test.py with 3000 steps trained model. Performance is not good enough. Need more training(2000000 according to author of crnn paper).
2.Tested training musicscore dataset, some warnings happened: No valid path.
02/23/2017 Update:
1.Tested training. Since CPU computation perforamnce is critical low, 1000 steps need ~2 hours.
2.Added test.py to apply model for testing picture.
02/20/2017 Update:
1.Added dataset.py for handle lmdb database.
2.Adjusted the architecture of directory.
02/17/2017 Update:
1.Added batch normalization to model.py
2.Added training.py for training dataset.
02/15/2017 Update:
1.Added model.py file.