peak_signal_to_noise_ratio.py

"""Peak Signal-to-Noise Ratio metric."""

import datasets
import numpy as np

from skimage.metrics import peak_signal_noise_ratio
from typing import Dict, Optional
import evaluate


_DESCRIPTION = """
Compute the Peak Signal-to-Noise Ratio (PSNR) for an image.

Please pay attention to the `data_range` parameter with floating-point images.
"""


_KWARGS_DESCRIPTION = """
Args:
    predictions (`list` of `np.array`): Predicted labels.
    references (`list` of `np.array`): Ground truth labels.
    sample_weight (`list` of `float`): Sample weights Defaults to None.
Returns:
    psnr (`float`): Peak Signal-to-Noise Ratio. The SSIM values are positive. Typical
    values for the PSNR in lossy image and video compression are between 30 and 50 dB,
    provided the bit depth is 8 bits, where higher is better. 
Examples:
    Example 1-A simple example
        >>> psnr = evaluate.load("jpxkqx/peak_signal_to_noise_ratio")
        >>> results = psnr.compute(references=[[0, 0], [-1, -1]], predictions=[[0, 1], [0, 0]])
        >>> print(results)
        {"Peak Signal-to-Noise Ratio": 23.010298856486173}
"""


_CITATION = """
@article{boulogne2014scikit,
  title={Scikit-image: Image processing in Python},
  author={Boulogne, Fran{\c{c}}ois and Warner, Joshua D and Neil Yager, Emmanuelle},
  journal={J. PeerJ},
  volume={2},
  pages={453},
  year={2014}
}
"""


@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
class PeakSignaltoNoiseRatio(evaluate.Metric):
    def _info(self):
        return evaluate.MetricInfo(
            description=_DESCRIPTION,
            citation=_CITATION,
            inputs_description=_KWARGS_DESCRIPTION,
            features=datasets.Features(self._get_feature_types()),
            reference_urls=["https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio"],
        )

    def _get_feature_types(self):
        if self.config_name == "multilist":
            return {
                # 1st Seq - num_samples, 2nd Seq - Height, 3rd Seq - Width
                "predictions": datasets.Sequence(
                    datasets.Sequence(datasets.Sequence(datasets.Value("float32")))
                ),
                "references": datasets.Sequence(
                    datasets.Sequence(datasets.Sequence(datasets.Value("float32")))
                ),
            }
        else:
            return {
                # 1st Seq - Height, 2rd Seq - Width
                "predictions": datasets.Sequence(
                    datasets.Sequence(datasets.Value("float32"))
                ),
                "references": datasets.Sequence(
                    datasets.Sequence(datasets.Value("float32"))
                ),
            }

    def _compute(
        self, 
        predictions, 
        references, 
        data_range: Optional[float] = None, 
        sample_weight=None,
    ) -> Dict[str, float]:
        samples = zip(np.array(predictions), np.array(references))
        psnrs = list(map(
            lambda args: peak_signal_noise_ratio(*args, data_range=data_range), samples
        ))
        return np.average(psnrs, weights=sample_weight)