image-quality: ValueError: the input array must have size 3 along `channel_axis`, got (80, 512)

~\Envs\Iris\lib\site-packages\imquality\brisque.py in score(image, kernel_size, sigma) 158 159 def score(image: PIL.Image.Image, kernel_size=7, sigma=7 / 6) -> float: –> 160 scaled_features = calculate_features(image, kernel_size, sigma) 161 return predict(scaled_features)

~\Envs\Iris\lib\site-packages\imquality\brisque.py in calculate_features(image, kernel_size, sigma) 128 129 def calculate_features(image: PIL.Image, kernel_size, sigma) -> numpy.ndarray: –> 130 brisque = Brisque(image, kernel_size=kernel_size, sigma=sigma) 131 # WARNING: The algorithm is very sensitive to rescale 132 # FIXME: this is empirically the best configuration; however, scikit-image warns about bi-quadratic implementation.

~\Envs\Iris\lib\site-packages\imquality\brisque.py in init(self, image, kernel_size, sigma) 43 ): 44 self.image = pil2ndarray(image) —> 45 self.image = skimage.color.rgb2gray(self.image) 46 self.kernel_size = kernel_size 47 self.sigma = sigma

~\Envs\Iris\lib\site-packages\skimage_shared\utils.py in fixed_func(*args, **kwargs) 336 337 if channel_axis is None: –> 338 return func(*args, **kwargs) 339 340 # TODO: convert scalars to a tuple in anticipation of eventually

~\Envs\Iris\lib\site-packages\skimage\color\colorconv.py in rgb2gray(rgb, channel_axis) 873 >>> img_gray = rgb2gray(img) 874 “”" –> 875 rgb = _prepare_colorarray(rgb) 876 coeffs = np.array([0.2125, 0.7154, 0.0721], dtype=rgb.dtype) 877 return rgb @ coeffs

~\Envs\Iris\lib\site-packages\skimage\color\colorconv.py in _prepare_colorarray(arr, force_copy, channel_axis) 138 msg = (f’the input array must have size 3 along channel_axis, ’ 139 f’got {arr.shape}') –> 140 raise ValueError(msg) 141 142 float_dtype = _supported_float_type(arr.dtype)

ValueError: the input array must have size 3 along channel_axis, got (80, 512)

About this issue

Original URL
State: open
Created 3 years ago
Reactions: 2
Comments: 15 (1 by maintainers)

Most upvoted comments

I found a solution to this issue. The problem is that in the calculate_features function in brisque.py, two Brisque objects get created.

Within the constructor of the Brique class, the function skimage.color.rgb2gray() gets called.

Back to the calculate_features method, the first Brique object that gets called, the input image is RGB; however, the second Brique object that gets called is already in grayscale. The skimage.color.rgb2gray() function throws an error that the input is already in grayscale as it expects 3 channels.

The solution was to add a condition around the call to skimage.color.rgb2gray() within the Brisque class.

Replace line 45 of `imquality/brisque.py’ with:

# Only convert image to grayscale if RGB
if self.image.shape[-1] == 3:
    self.image = skimage.color.rgb2gray(self.image)

+24

njay225 on Jan 12, 2022

@ocampor I believe a new pypi package is needed that includes the fix https://github.com/ocampor/image-quality/pull/43. Otherwise the fix won’t be generally available…

HodeiG on Jul 12, 2022

@BMaser did you install the dependency directly from github master or did you use pip? I believe the latter won’t work because a new package hasn’t been rebuilt since Feb 2021: https://pypi.org/project/image-quality/#files

HodeiG on Apr 24, 2022