Shaohui Ruan

@zunair11 the code need about 1.5G memory.

@guddulrk `python setup.py build_ext --inplace`

@guddulrk @Guo-king if your are using windows, use nms and bbox implemented in pure python instead, you have to modify some import

@aa642531 you can download the testing scripts from [here](http://rrc.cvc.uab.es/?ch=1&com=introduction)

@YangJiao1996 in demo.py the detection result has been saved to txt file, you can download the evaluation scripts from [here](http://rrc.cvc.uab.es/?ch=2&com=mymethods&task=1) if you wanna test on icdar11,13 or 15

@kofzhoudong hi, put them in utils/bbox/

@guddulrk hi, I have update the readme, check it for more detail about the training data

@honghande download the model from release

@chengjunjiecn I did not test on cpu. with gpu, the speed can be reduce to 0.4s per image

@zhangxiaohuixuhao [build_color_map](https://github.com/eragonruan/refinenet-image-segmentation/blob/master/build_color_map.py) is used to build the mapping relationship between class and color. This is used for visualization, nothing to do with the model