white

多谢回复。 主要是读取coco数据集地方做了改动，网络结构和loss都没变化。如果可以的话，我把主要代码发到[email protected]，还请帮忙看看。 多谢啦

I don't think there is really a trained model you could save. Actually, only the training data are saved, as well as the KnnDtw object itself.

> Thank you for your praise! I have not tried resnet50 yet(`network/resnet` is an incomplete version), if you are going to do this, try importing resnet into textnet.py. In the...

> have you solved it ? not yet. However, vgg16 in this case still works fine.

> > > @white2018 https://github.com/Longqi-S/keras_cpn/blob/master/postprocessing/mptest.py this is predict script. That is an evaluation program. what I mean is an end-to-end predicting program with the image as input and the keypoint...

It looks like it happens for the training images, in which the stroke width varies. Does it mean this algorithm only works fine for the fixed stroke width?

> 英文单词之间是没有空格的。识别出来的英文段落，是混在一起的。作者有什么办法fix吗？ 这个要把空格也作为字符集试试看呢？还需要准备相应的数据集