InvalidArgumentError Related to the Shape of Tensors When Utilizing tf.image.random

I'm working with a GAN using tensorflow.

inp, re = load(str(PATH / 'train/100.jpg'))
    # Casting to int for matplotlib to display the images
    plt.figure()
    plt.imshow(inp / 255.0)
    plt.figure()
    plt.imshow(re / 255.0)

When running the above code, the following error occurs:

InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-17-abf50abadd26> in <module>
----> 1 inp, re = load(str(PATH / 'train/100.jpg'))
      2 # Casting to int for matplotlib to display the images
      3 plt.figure()
      4 plt.imshow(inp / 255.0)
      5 plt.figure()

<ipython-input-16-e73db7456b4c> in load(image_file)
     16     #crop and resize
     17     input_image = cv2.resize(input_image,(width1 * 10,height1 * 10))
---> 18     input_image = tf.image.random_crop(input_image,(width1,height1))
     19     real_image = cv2.resize(real_image,(width1,height1))
     20 

/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/util/dispatch.py in wrapper(*args, **kwargs)
    204     """Call target, and fall back on dispatchers if there is a TypeError."""
    205     try:
--> 206       return target(*args, **kwargs)
    207     except (TypeError, ValueError):
    208       # Note: convert_to_eager_tensor currently raises a ValueError, not a

/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/ops/random_ops.py in random_crop(value, size, seed, name)
    400     shape = array_ops.shape(value)
    401     check = control_flow_ops.Assert(
--> 402         math_ops.reduce_all(shape >= size),
    403         ["Need value.shape >= size, got ", shape, size],
    404         summarize=1000)

/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/ops/math_ops.py in wrapper(x, y, *args, **kwargs)
   1815   def wrapper(x, y, *args, **kwargs):
   1816     x, y = maybe_promote_tensors(x, y, force_same_dtype=False)
-> 1817     return fn(x, y, *args, **kwargs)
   1818   return tf_decorator.make_decorator(fn, wrapper)
   1819 

/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/ops/gen_math_ops.py in greater_equal(x, y, name)
   4030       return _result
   4031     except _core._NotOkStatusException as e:
-> 4032       _ops.raise_from_not_ok_status(e, name)
   4033     except _core._FallbackException:
   4034       pass

/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/framework/ops.py in raise_from_not_ok_status(e, name)
   6939   message = e.message   (" name: "   name if name is not None else "")
   6940   # pylint: disable=protected-access
-> 6941   six.raise_from(core._status_to_exception(e.code, message), None)
   6942   # pylint: enable=protected-access
   6943 

/opt/anaconda3/lib/python3.8/site-packages/six.py in raise_from(value, from_value)

InvalidArgumentError: Incompatible shapes: [3] vs. [2] [Op:GreaterEqual]

To my understanding, this means that a tensor is the incorrect shape, but how to I know what is causing this difference, or how to fix it?

It is certainly worth noting that the load() function is defined by:

def load(image_file):
    #Read and decode an image file to a uint8 tensor
    image = tf.io.read_file(image_file)
    image = tf.io.decode_jpeg(image)

    #conversion to numPy array
    image = np.array(image)
    
    #seperate
    input_image = image
    real_image = image
    
    #crop and resize
    input_image = cv2.resize(input_image,(width1 * 10,height1 * 10))
    input_image = tf.image.random_crop(input_image,(width1,height1))
    real_image = cv2.resize(real_image,(width1,height1))
    
    #convert to float32
    input_image = tf.cast(input_image,tf.float32) #51,34
    real_image = tf.cast(real_image,tf.float32) #510,340
    
    return input_image, real_image

When tf.image.random_crop is removed, the error is not encountered, however the images are not properly cropped, of course. I'm not entirely sure why this happens.

CodePudding user response：

inp, re = load(str(PATH/'train/100.jpg'))

This is where the error comes from initially. It looks like you forgot to define the image before this command.

CodePudding user response：

The error was related to the syntax required for tf.image.random_crop(). While I used tf.image.random_crop(image, (width,height)), the correct syntax is tf.image.random_crop(image, size = [width,height,3]) 3 Is used here as the image is in color. The syntax for this was rather unclear from sources I had previously consulted; thusly, I hope this answer helps someone.