search

MaybeShewill-CV / CRNN_Tensorflow

Convolutional Recurrent Neural Networks(CRNN) for Scene Text Recognition
MIT License
1.03k stars 388 forks source link

@LXLun cv2.resize(image, dsize=tuple(CFG.ARCH.INPUT_SIZE), interpolation=cv2.IMREAD_GRAYSCALE) 建议你多看点基础只是 IMREAD_GRAYSCALE不是插值方法 #326

Closed LXLun closed 5 years ago

LXLun commented 5 years ago

@LXLun cv2.resize(image, dsize=tuple(CFG.ARCH.INPUT_SIZE), interpolation=cv2.IMREAD_GRAYSCALE) 建议你多看点基础只是 IMREAD_GRAYSCALE不是插值方法

_Originally posted by @MaybeShewill-CV in https://github.com/MaybeShewill-CV/CRNN_Tensorflow/issues/323#issuecomment-515824060_

这是我前面写错了位置，我生成数据用的是 image = cv2.imread(sample_path, cv2.IMREAD_GRAYSCALE) image = cv2.resize(image, dsize=tuple(CFG.ARCH.INPUT_SIZE), interpolation=cv2.INTER_LINEAR)

这不是这个错误的原因........

MaybeShewill-CV commented 5 years ago

@LXLun 单通道当然不能用现有的模型直接训练 需要修改一些tensor的shape这个你自己看代码改吧:)

LXLun commented 5 years ago

可以指点下大概的需要修改那个地方么

MaybeShewill-CV commented 5 years ago

@LXLun 报错提示已经很清楚了 哪里的维度不匹配你就改相应的地方:)

LXLun commented 5 years ago

问一下，你的配置一张图片大概要解析耗时多长时间