The width and Height is misplaced

[MisayaZ]

img_size = [94, 24]
inputs = tf.placeholder(tf.float32, shape=(b, img_size[0], img_size[1], num_channels))
however, the shape order should be [batchsize, height, width, channel]

[lyl8213]

I rotate the images when reading the image, as follows:
images = np.transpose(images, axes=[0, 2, 1, 3])
The image size is setting according to the paper.

[MisayaZ]

@lyl8213 Why do you rotate the images?
what is the benefits of this?

[lyl8213]

Sorry,I didn't figure it out. The paper's input is 94*24, so i have to rotate the image. Moreover, I found some plate recognition programs in github also rotate the image.