Minghui Liao

@zhongdajian I think you should not use the distributed mode for testing. Generally, it is only used in training.

@LittleCurry This is related to the Pytorch version. You may need to modify some code if you use a higher Pytorch version.

@Rosesor The caffe2 version is an implementation of the ECCV paper while this repo is an implementation of the TPAMI paper.

@vishaljainai Both of them should be okay.

You can pre-train the model with SynthText.

@Shualite I have provided the converted labels of SynthText [here](https://github.com/MhLiao/MaskTextSpotter#datasets).

@ChChwang I am now working for the single image inference demo and the evaluation code.

@yuzhouzhili [Here](https://github.com/MhLiao/MaskTextSpotter#datasets) are some examples for the labels.

Thanks for your contribution! I am checking the pull request. It seems that you modified the indentations so it is hard to see the actual changes of the code since...

@iamrishab It seems that the char maps are wrongly localized with the current PR.