sirius-ai

1) Please try to use little learning rate firstly. 2) Please check your training datasets whether have unclean data in it.

LPRNet本来就是设计输入大小为94x24的，所以你修改了输入大小后当然会报错了！

Yes, it depend on the data distribution how like in synthetic data and real images .

相同单词如果中间没有空白符进行隔开就会被合并的，看样子还是算法没有训练好，试试增加这方面的训练数据看看。

可以转的，也可以试试libtorch.

1. if it is not chinese license plate and license plate lenth equal to 8, you should delete check(label) funtcion in LPRDataLoader.__getitem__; 2. you should modify the list of CHARS...

@2017TJM please let me know how to reproduce this issue, thanks!

我没有看到OpenCV里有开放LPRNet啊，可以告知在哪里不？

most importance api in verification.py is evaluate，and its parameters annotaion as below: emb_array：a list, model output embedding vector. issame_list：a list, tell the odd and even elements of emb_array are same...

like this: diff = np.subtract(embed1, embed2) dist = np.sum(np.square(diff), 1)