NaturalImageQualityEvaluator¶

class mmeval.metrics.NaturalImageQualityEvaluator(crop_border: int = 0, input_order: str = 'CHW', convert_to: str = 'gray', channel_order: str = 'rgb', **kwargs)[source]¶

Calculate Natural Image Quality Evaluator(NIQE) metric.

Ref: Making a “Completely Blind” Image Quality Analyzer. This implementation could produce almost the same results as the official MATLAB codes: http://live.ece.utexas.edu/research/quality/niqe_release.zip

Parameters

crop_border (int) – Cropped pixels in each edges of an image. These pixels are not involved in the NIQE calculation. Defaults to 0.
input_order (str) – Whether the input order is ‘HWC’ or ‘CHW’. Defaults to ‘CHW’.
convert_to (str) – Convert the images to other color models. Options are ‘y’ and ‘gray’. Defaults to ‘gray’.
channel_order (str) – The channel order of image. Defaults to ‘rgb’.
**kwargs – Keyword parameters passed to BaseMetric.

Examples

>>> from mmeval import NaturalImageQualityEvaluator
>>> import numpy as np
>>>
>>> niqe = NaturalImageQualityEvaluator()
>>> preds = np.random.randint(0, 255, size=(3, 32, 32))
>>> niqe(preds)  
{'niqe': ...}

add(predictions: Sequence[numpy.ndarray], channel_order: Optional[str] = None) → None[source]¶

Add NIQE score of batch to self._results

Parameters

predictions (Sequence[np.ndarray]) – Predictions of the model. Each prediction should be a 4D (as a video, not batches of images) or 3D (as an image) array.
channel_order (Optional[str]) – The channel order of the input samples. If not passed, will set as self.channel_order. Defaults to None.

compute_feature(block: numpy.ndarray) → List[source]¶

Compute features.

Parameters: block (np.ndarray) – 2D Image block.
Returns: Features with length of 18.
Return type: List

compute_metric(results: List[numpy.float64]) → Dict[str, float][source]¶

Compute the NaturalImageQualityEvaluator metric.

This method would be invoked in BaseMetric.compute after distributed synchronization.

Parameters: results (List[np.float64]) – A list that consisting the NIQE score. This list has already been synced across all ranks.
Returns: The computed NIQE metric.
Return type: Dict[str, float]

compute_niqe(prediction: numpy.ndarray, channel_order: str, mu_pris_param: numpy.ndarray, cov_pris_param: numpy.ndarray, gaussian_window: numpy.ndarray, block_size_h: int = 96, block_size_w: int = 96) → numpy.float64[source]¶

Calculate NIQE (Natural Image Quality Evaluator) metric. We use the official params estimated from the pristine dataset. We use the recommended block size (96, 96) without overlaps.

Note that we do not include block overlap height and width, since they are always 0 in the official implementation.

For good performance, it is advisable by the official implementation to divide the distorted image into the same size patched as used for the construction of multivariate Gaussian model.

Parameters

prediction (np.ndarray) – Input image whose quality to be computed. Range [0, 255] with float type.
channel_order (str) – The channel order of image.
mu_pris_param (np.ndarray) – Mean of a pre-defined multivariate Gaussian model calculated on the pristine dataset.
cov_pris_param (np.ndarray) – Covariance of a pre-defined multivariate Gaussian model calculated on the pristine dataset.
gaussian_window (ndarray) – A 7x7 Gaussian window used for smoothing the image.
block_size_h (int) – Height of the blocks in to which image is divided. Defaults to 96.
block_size_w (int) – Width of the blocks in to which image is divided. Defaults to 96.

Returns

NIQE result.

Return type

np.float64

estimate_aggd_param(block: numpy.ndarray) → Tuple[source]¶

Estimate AGGD (Asymmetric Generalized Gaussian Distribution) parameters.

Parameters: block (np.ndarray) – 2D Image block.
Returns: alpha(float), beta_l(float) and beta_r(float) for the AGGD distribution (Estimating parameters in Equation 7 in the paper).
Return type: Tuple

get_size_from_scale(input_size: Tuple, scale_factor: List[float]) → List[int][source]¶

Get the output size given input size and scale factor.

Parameters

input_size (Tuple) – The size of the input image.
scale_factor (List[float]) – The resize factor.

Returns

The size of the output image.

Return type

List[int]

get_weights_indices(input_length: int, output_length: int, scale: float, kernel: Callable, kernel_width: float) → Tuple[List[numpy.ndarray], List[numpy.ndarray]][source]¶

Get weights and indices for interpolation.

Parameters

input_length (int) – Length of the input sequence.
output_length (int) – Length of the output sequence.
scale (float) – Scale factor.
kernel (Callable) – The kernel used for resizing.
kernel_width (float) – The width of the kernel.

Returns

The weights and the indices for interpolation.

Return type

Tuple[List[np.ndarray], List[np.ndarray]]

matlab_resize(img: numpy.ndarray, scale: float) → numpy.ndarray[source]¶

Resize an image to the required size.

Parameters

img (np.ndarray) – The original image.
scale (float) – The scale factor of the resize operation.

Returns

The resized image.

Return type

np.ndarray

resize_along_dim(img_in: numpy.ndarray, weights: numpy.ndarray, indices: numpy.ndarray, dim: int) → numpy.ndarray[source]¶

Resize along a specific dimension.

Parameters

img_in (np.ndarray) – The input image.
weights (np.ndarray) – The weights used for interpolation, computed from [get_weights_indices].
indices (np.ndarray) – The indices used for interpolation, computed from [get_weights_indices].
dim (int) – Which dimension to undergo interpolation.

Returns

Interpolated (along one dimension) image.

Return type

np.ndarray