Error when using tensor flow grayscale conversion in certain images

When I tried to write a program that reads and processes images from Google Drive, an error occurred in a specific image.I was able to process other images without any problems, but when I run them with this image, the following error occurs.It seems that there is no problem with loading and tensorizing.
The language used is python and the running environment is Google collaboration.
The tensor flow version is 2.0.0.

How can I resolve this error?

当 I am not familiar with this site, so I would appreciate it if you could point it out if there is anything wrong with how to ask questions or write them down.

InvalidArgumentError Traceback (most recent call last)
<ipython-input-111-242afefc9006>in<module>()
  1 pre_data = tf.image.resize (pre_data, [500,500])
---->2pre_data=tf.image.rgb_to_grayscale(pre_data)
  3 pre_data = np.asarray(pre_data)
  4 pre_data=np.squeeze(pre_data)
  5pre_data.shape
 -----------------------------------------------------------------------------------------------------------------------------------------
/usr/local/lib/python 3.6/dist-packages/six.py in raise_from (value, from_value)

InvalidArgumentError: Matrix size-incompatible: In[0]: [250000,4], In[1]: [3,1] [Op: MatMul] name: rgb_to_grayscale/Tensordot/MatMul/

< print() is omitted.Also, in the original Colaboratory program, the cells are filled in and written (please let me know if there is a good way to write)

import tensorflow as tf
import numpy as np

path="/content/drive/My Drive/for_MikuChecker"

img=np.asarray(image.open(path+"/run_check/yes/48911351_p0.png"))
img=tf.convert_to_tensor(img)#TensorShape([992,622,4])
size = if h<welse w#w,h = (622,992)
pre_data=tf.image.crop_to_bounding_box(img, int((h-size)/2), int(w-size)/2), size, size)
pre_data=tf.image.resize(pre_data, [500,500])
pre_data=tf.image.rgb_to_grayscale(pre_data)
pre_data=np.asarray(pre_data)
pre_data=np.squeeze(pre_data)
pre_data.shape

Where the error occurred

pre_data=tf.image.rgb_to_grayscale(pre_data)

Image used

This image is reprinted from (Head Office)

python python3 tensorflow image google-colaboratory

2022-09-29 22:37

1 Answers

According to the Tensorflow document, the tf.image.rgb_to_grayscale function requires the last dimension to be 3.

https://www.tensorflow.org/api_docs/python/tf/image/rgb_to_grayscale

Args:

    images: The RGB tensor to convert. Last dimension must have size 3 and should contain RGB values.

But if you look at the source code in question,

img=tf.convert_to_tensor(img)#TensorShape([992,622,4])

The last dimension is 4, so I think there is an error.The image in question appears to be an RGBA image, not an RGB image. If you create a tensor using only three RGB channels using the above code as shown below, you will be able to avoid the time being.(If alpha channel is set, we need to deal with it separately.)

img=tf.convert_to_tensor(img[:,:,:3])

2022-09-29 22:37

If you have any answers or tips

Popular Tags

python x 4647

android x 1593

java x 1494

javascript x 1427

c x 927

c++ x 878

ruby-on-rails x 696

php x 692

python3 x 685

html x 656