horizon-metrics.py

# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import evaluate
import datasets
import numpy as np

from seametrics.horizon.utils import *

_CITATION = """\
@InProceedings{huggingface:module,
title = {Horizon Metrics},
authors={huggingface, Inc.},
year={2024}
}
"""

# TODO: Add description of the module here
_DESCRIPTION = """\
This metric is intended to calculate horizon prediction metrics."""

# TODO: Add description of the arguments of the module here
_KWARGS_DESCRIPTION = """
Calculates how good are predictions given some references, using certain scores
Args:
    predictions: list of predictions for each image. Each prediction
        should be a nested array like this:
        - [[x1, y1], [x2, y2]]

    references: list of references for each image. Each reference
        should be a nested array like this:
        - [[x1, y1], [x2, y2]]
Returns:
    dict containing following metrics:
    'average_slope_error': Measures the average difference in slope between the predicted and ground truth horizon.
    'average_midpoint_error': Calculates the average difference in midpoint position between the predicted and ground truth horizon.
    'stddev_slope_error': Indicates the variability of errors in slope between the predicted and ground truth horizon.
    'stddev_midpoint_error': Quantifies the variability of errors in midpoint position between the predicted and ground truth horizon.
    'max_slope_error': Represents the maximum difference in slope between the predicted and ground truth horizon.
    'max_midpoint_error': Indicates the maximum difference in midpoint position between the predicted and ground truth horizon.
    'num_slope_error_jumps': Calculates the differences between errors in successive frames for the slope. It then counts the number of jumps in these errors by comparing the absolute differences to a specified threshold.
    'num_midpoint_error_jumps': Calculates the differences between errors in successive frames for the midpoint. It then counts the number of jumps in these errors by comparing the absolute differences to a specified threshold.

Examples:
    >>> ground_truth_points = [[[0.0, 0.5384765625], [1.0, 0.4931640625]],
                       [[0.0, 0.53796875], [1.0, 0.4928515625]],
                       [[0.0, 0.5374609375], [1.0, 0.4925390625]],
                       [[0.0, 0.536953125], [1.0, 0.4922265625]],
                       [[0.0, 0.5364453125], [1.0, 0.4919140625]]]

    >>> prediction_points = [[[0.0, 0.5428930956049597], [1.0, 0.4642497615378973]],
                     [[0.0, 0.5428930956049597], [1.0, 0.4642497615378973]],
                     [[0.0, 0.523573113510805], [1.0, 0.47642688648919496]],
                     [[0.0, 0.5200016849393765], [1.0, 0.4728554579177664]],
                     [[0.0, 0.523573113510805], [1.0, 0.47642688648919496]]]


    >>> module = evaluate.load("SEA-AI/horizon-metrics", roll_threshold=0.5, pitch_threshold=0.1, vertical_fov_degrees=25.6, height=512)
    >>> module.add(predictions=ground_truth_points, references=prediction_points)
    >>> module.compute()
    >>> {'average_slope_error': 0.014823194839790999,
         'average_midpoint_error': 0.014285714285714301,
         'stddev_slope_error': 0.01519178791378349,
         'stddev_midpoint_error': 0.0022661781575342445,
         'max_slope_error': 0.033526146567062376,
         'max_midpoint_error': 0.018161272321428612,
         'num_slope_error_jumps': 1,
         'num_midpoint_error_jumps': 1}
    """


@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION,
                                                _KWARGS_DESCRIPTION)
class HorizonMetrics(evaluate.Metric):
    """
    HorizonMetrics is a metric class that calculates horizon prediction metrics.

    Args:
        roll_threshold (float, optional): Threshold for roll angle. Defaults to 0.5.
        pitch_threshold (float, optional): Threshold for pitch angle. Defaults to 0.1.
        vertical_fov_degrees (float, optional): Vertical field of view in degrees. Defaults to 25.6.
        **kwargs: Additional keyword arguments.

    Attributes:

        slope_threshold (float): Threshold for slope calculated from roll threshold.
        midpoint_threshold (float): Threshold for midpoint calculated from pitch threshold.
        predictions (list): List of predicted horizons.
        ground_truth_det (list): List of ground truth horizons.
        slope_error_list (list): List of slope errors.
        midpoint_error_list (list): List of midpoint errors.

    Methods:

        _info(): Returns the metric information.
        add(predictions, references, **kwargs): Updates the predictions and ground truth detections.
        _compute(predictions, references, **kwargs): Computes the horizon error across the sequence.
    """

    def __init__(self,
                 roll_threshold=0.5,
                 pitch_threshold=0.1,
                 vertical_fov_degrees=25.6,
                 height=512,
                 **kwargs):

        super().__init__(**kwargs)

        self.slope_threshold = roll_to_slope(roll_threshold)
        self.midpoint_threshold = pitch_to_midpoint(pitch_threshold,
                                                    vertical_fov_degrees)
        self.predictions = None
        self.ground_truth_det = None
        self.slope_error_list = []
        self.midpoint_error_list = []
        self.height = height
        self.vertical_fov_degrees = vertical_fov_degrees

    def _info(self):
        """
        Returns the metric information.

        Returns:
            MetricInfo: The metric information.
        """
        return evaluate.MetricInfo(
            # This is the description that will appear on the modules page.
            module_type="metric",
            description=_DESCRIPTION,
            citation=_CITATION,
            inputs_description=_KWARGS_DESCRIPTION,
            # This defines the format of each prediction and reference
            features=datasets.Features({
                'predictions':
                datasets.Sequence(
                    datasets.Sequence(
                        datasets.Sequence(datasets.Value("float")))),
                'references':
                datasets.Sequence(
                    datasets.Sequence(
                        datasets.Sequence(datasets.Value("float")))),
            }),
            codebase_urls=["http://github.com/path/to/codebase/of/new_module"])

    def add(self, *, predictions, references, **kwargs):
        """
        Updates the predictions and ground truth detections.

        Parameters:
            predictions (list): List of predicted horizons.
            references (list): List of ground truth horizons.
            **kwargs: Additional keyword arguments.
        """
        super(evaluate.Metric, self).add(prediction=predictions,
                                         references=references,
                                         **kwargs)

        self.predictions = predictions
        self.ground_truth_det = references

    def _compute(self, *, predictions, references, **kwargs):
        """
        Computes the horizon error across the sequence.

        Returns:
            float: The computed horizon error.
        """

        # calculate erros and store values in slope_error_list and midpoint_error_list
        for annotated_horizon, proposed_horizon in zip(self.ground_truth_det,
                                                       self.predictions):

            if annotated_horizon is None or proposed_horizon is None:
                continue
            slope_error, midpoint_error = calculate_horizon_error(
                annotated_horizon, proposed_horizon)
            self.slope_error_list.append(slope_error)
            self.midpoint_error_list.append(midpoint_error)

        # calculate slope errors, midpoint errors and jumps
        result = calculate_horizon_error_across_sequence(
            self.slope_error_list, self.midpoint_error_list,
            self.slope_threshold, self.midpoint_threshold,
            self.vertical_fov_degrees, self.height)

        # calulcate detection rate
        detected_horizon_count = len(
            self.predictions) - self.predictions.count(None)
        detected_gt_count = len(
            self.ground_truth_det) - self.ground_truth_det.count(None)

        detection_rate = detected_horizon_count / detected_gt_count
        result['detection_rate'] = detection_rate

        return result