1. Here i choose a small dataset from [Synthetic_Chinese_String_Dataset](https://github.com/chenjun2hao/caffe_ocr), about 270000+ images for training, 20000 images for testing.
download the image data from [Baidu](https://pan.baidu.com/s/1hIurFJ73XbzL-QG4V-oe0w)
there are some details for attention
2. the train_list.txt and test_list.txt are created as the follow form:
```
# path/to/image_name.jpg label
...
...
@@ -66,9 +65,3 @@ then you can see in the terminel as follow:
- [ ] change LSTM to Conv1D, it can greatly accelerate the inference
- [ ] to support images of different widths
- [ ] change the cnn bone model with inception net, densenet
