SumAbsoluteDifferences¶

class mmeval.metrics.SumAbsoluteDifferences(norm_const: int = 1000, **kwargs)[source]¶

Sum of Absolute Differences metric for image.

This metric computes per-pixel absolute difference and sum across all pixels. i.e. sum(abs(a-b)) / norm_const

Parameters

norm_const (int) – Divide the result to reduce its magnitude. Default to 1000.
**kwargs – Keyword parameters passed to BaseMetric.

Note

The current implementation assumes the image a numpy array with pixel values ranging from 0 to 255.

Examples

>>> from mmeval import SumAbsoluteDifferences as SAD
>>> import numpy as np
>>>
>>> sad = SAD()
>>> prediction = np.zeros((32, 32), dtype=np.uint8)
>>> groundtruth = np.ones((32, 32), dtype=np.uint8) * 255
>>> sad(prediction, groundtruth)  
{'sad': ...}

add(predictions: Sequence[numpy.ndarray], groundtruths: Sequence[numpy.ndarray]) → None[source]¶

Add SumAbsoluteDifferences score of batch to self._results

Parameters

predictions (Sequence[np.ndarray]) – Sequence of predicted image.
groundtruths (Sequence[np.ndarray]) – Sequence of groundtruth image.

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

Compute the SumAbsoluteDifferences metric.

Parameters: results (List) – A list that consisting the SumAbsoluteDifferences score. This list has already been synced across all ranks.
Returns: The computed SumAbsoluteDifferences metric. The keys are the names of the metrics, and the values are corresponding results.
Return type: Dict[str, float]