Pay20Y
Here is the Google Drive link https://drive.google.com/file/d/1JSCCXYlpCiSuLj41jYhPQDPU5PGVo3UF/view?usp=sharing
Directly only trained on ICDAR13 is not suitable, since there are too few images. You can trained on SynthText and Synth90K which can be found [here](https://github.com/ayumiymk/aster.pytorch/issues/17#issuecomment-543757633). And you can finetune...
I'm so sorry. I checked it again. The accuracy 91.8 is achieved on ICDAR13-857 not 1015.
Oh yes I will, maybe tomorrow. And I will update how to use.
I'm not sure. And the results stated in readme is from [this paper](https://arxiv.org/pdf/1904.01375v1.pdf)
I have ran the this one, it can successfully train and test but I didn't test it on those benchmark datasets.
I guess @delveintodetail means _teacher forcing_. However, in predict.py _teacher forcing_ is not adopted, so I don't think there are bugs in predict.py
不好意思，我只是问一下这个问题，并没有实现训练代码
Hey I have the same question with you. How to make the affine matrix computed from OpenCV work? Did you solve it? Thanks! @ @Justdjent
I solved it. Sorry to bother you.