xiaxiaoling

@BrownSweater 我也在看识别部分的代码，改了一点可以跑起来但是效果不好，可以交流下吗

@Wovchena you can see my issues https://github.com/jiangxiluning/FOTS.PyTorch/issues/29 now i can train in synth data, but i don't kown the result, training need a lot of time.

@northeastsquare @MiZhangWhuer @jiangxiluning I have some question about recognition loss, when validation, the model calculate the recognition loss use the predict boxes (in code), then the predict boxes cant't map...

@northeastsquare “RuntimeError ： targetlengths mus be of size batch size”

@Pay20Y why it recognise so slow? in paper, it says the time will less than detect plus recognise, but it more than detect plus recogise actually (when 200 text).

@insightcs hi, if I want to use this soft mask label, need I add this code to the project? I can't find about soft mask label in the project.

@Govan111 我现在结果都正常，你可以试试debug到计算acc的地方，看看结果到底预测成什么了

@Govan111 我是加入了计算编辑距离，所以写了个回调函数计算acc和编辑距离，keras自带的计算acc地方就debug不到了

@Govan111 我使用学习率0.001在多个数据集上（自己准备的，打印体手写体都有）都能达到0.8-0.9多的acc，有可能是你数据集问题，也有可能是你计算acc的问题，因为我觉得这个代码计算acc有些问题，使用loss为0记作一个正确的样本的

https://github.com/xieyufei1993/FOTS/issues/3 @poppy007 it's a python package you can download,