IQM_Vis.metrics package

Subpackages

Submodules

IQM_Vis.metrics.metric_utils module

Helper functions for metric computation

IQM_Vis.metrics.non_perceptual module

class IQM_Vis.metrics.non_perceptual.MAE(return_image=False)[source]

Bases: object

Mean Absolute Error between two images. Images must have the same

dimensions

Parameters

return_image (bool) – Whether to return the image (Defaults to False which will return a scalar value)

__call__(im_ref, im_comp, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

MAE (scalar if return_image is False, image if

return_image is True)

Return type

score (np.array)

class IQM_Vis.metrics.non_perceptual.MSE(return_image=False)[source]

Bases: object

Mean Squared Error between two images. Images must have the same

dimensions

Parameters

return_image (bool) – Whether to return the image (Defaults to False which will return a scalar value)

__call__(im_ref, im_comp, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

MSE (scalar if return_image is False, image if

return_image is True)

Return type

score (np.array)

class IQM_Vis.metrics.non_perceptual.RMSE(return_image=False)[source]

Bases: object

Root Mean Squared Error between two images. Images must have the same

dimensions

Parameters

return_image (bool) – Whether to return the image (Defaults to False which will return a scalar value)

__call__(im_ref, im_comp, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

RMSE (scalar if return_image is False, image if

return_image is True)

Return type

score (np.array)

class IQM_Vis.metrics.non_perceptual.one_over_PSNR[source]

Bases: object

Peak signal to noise ratio - https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio The score given is normalised between 0, 1 by taking 1/PSNR

Parameters

return_image (bool) – Whether to return the image (Defaults to False which will return a scalar value)

__call__(im_ref, im_comp, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

PSNR between the images

Return type

score (np.array)

IQM_Vis.metrics.perceptual_DL module

class IQM_Vis.metrics.perceptual_DL.DISTS[source]

Bases: object

Deep Image Structure and Texture Similarity (DISTS) Metric. Uses the code from https://github.com/dingkeyan93/DISTS. Uses the PyTorch backend. It is robust to texture variance (e.g., evaluating the images generated by GANs) and mild geometric transformations (e.g., evaluating the image pairs that are not strictly point-by-point aligned).

__call__(im_ref, im_comp, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

DISTS score

Return type

score (np.array)

class IQM_Vis.metrics.perceptual_DL.LPIPS(network='alex', reduction='mean')[source]

Bases: object

Learned Perceptual Image Patch Similarity between two images.

Images must have the same dimensions.

Parameters

  • network (str) – Pretrained network to use. Choose between ‘alex’, ‘vgg’ or ‘squeeze’. (Defaults to ‘alex’)

  • reduction (str) – How to reduce over the batch dimension. Choose between ‘sum’ or ‘mean’. (Defaults to ‘mean’)

__call__(im_ref, im_comp, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

LPIPS score

Return type

score (np.array)

IQM_Vis.metrics.perceptual_trad module

class IQM_Vis.metrics.perceptual_trad.MS_SSIM(return_image=False)[source]

Bases: object

Multi-Scale Structural Similarity Index Measure between two images.

Images must have the same dimensions. Score given is 1 - MS_SSIM to give the loss/dissimilarity. Note that images of small size, below 180 pixels will have their kernel size reduced for compatability with the 4 downsizing operations.

Parameters

return_image (bool) – Whether to return the image (Defaults to False which will return a scalar value)

__call__(im_ref, im_comp, sigma=1.5, k1=0.01, k2=0.03, mssim_kernel_size=11, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

1-SSIM (scalar if return_image is False, image if

return_image is True)

Return type

score (np.array)

class IQM_Vis.metrics.perceptual_trad.NLPD[source]

Bases: object

Normalised Laplacian pyramid Proposed by Valero Laparra et al. https://www.uv.es/lapeva/papers/2016_HVEI.pdf . NLPD is an image quality metric based on the transformations associated with the early visual system: local luminance subtraction and local gain control.

__call__(im_ref, im_comp, nlpd_k=1, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

NLPD score

Return type

score (np.array)

class IQM_Vis.metrics.perceptual_trad.SSIM(return_image=False)[source]

Bases: object

Structural Similarity Index Measure between two images. Images must have

the same dimensions. Score given is 1 - SSIM to give the loss/dissimilarity

Parameters

return_image (bool) – Whether to return the image (Defaults to False which will return a scalar value)

__call__(im_ref, im_comp, sigma=1.5, k1=0.01, k2=0.03, ssim_kernel_size=11, **kwargs)[source]

When an instance is called

Parameters

  • im_ref (np.array) – Reference image

  • im_comp (np.array) – Comparison image

  • **kwargs – Arbitrary keyword arguments

Returns

1-SSIM (scalar if return_image is False, image if

return_image is True)

Return type

score (np.array)

Module contents

IQM_Vis.metrics.get_all_IQM_params()[source]

Get all available IQMs parameters

Returns

  • all_params (dict)

IQM_Vis.metrics.get_all_metric_images()[source]

Get all available IQMs provided by IQM_Vis that return an image

Returns

  • all_metrics (dict)

IQM_Vis.metrics.get_all_metrics()[source]

Get all available IQMs provided by IQM_Vis

Returns

  • all_metrics (dict)