ANN-route-predition/pyrate/tests/sense/vision/test_obstacle_locator.py

104 lines
4.0 KiB
Python

"""This test suite evaluates and tests behavior of the ``ObstacleLocator`` class"""
# Standard library
from csv import reader
from math import radians
from pathlib import Path
# Typing
from typing import cast
from typing import Tuple
# Testing
from unittest import TestCase
# Scientific
from cv2 import imread
# Module under test
from pyrate.sense.vision.image_line import ImageLine
from pyrate.sense.vision.obstacle_locator import ObstacleLocator
PATH_TO_DATASET = Path(__file__).parent / "resources" / "testing_dataset_successful"
DATASET_IMAGES_PATHS = sorted(list((PATH_TO_DATASET / "testims").glob("*.jpg")))
DATASET_ANNOTATIONS_PATHS = sorted(list((PATH_TO_DATASET / "annotations").glob("*.txt")))
PATH_TO_FAILING = (
Path(__file__).parent / "resources" / "testing_dataset_no_horizon" / "testims" / "Preprocessed_test_0.jpg"
)
IMAGE_HEIGHT, IMAGE_WIDTH = imread(PATH_TO_FAILING.as_posix()).shape[:2]
class TestObstacleLocator(TestCase):
"""Test for correct predictions made by ``ObstacleLocator``"""
@staticmethod
def parse_annotation(file_path: str, obstacle_locator: ObstacleLocator) -> Tuple[ImageLine, float]:
"""Helper function to parse the ground truth labels from the dataset.
Args:
file_path: Label file path
obstacle_locator: the ObstacleLocator that returns the ImageLine that should be
compared to the returned ImageLine of this function
Returns:
ImageLine as described in the annotation, angle read from annotation
(for testing correct angle calculation)
"""
with open(file_path, "rt", encoding="UTF-8") as label_file:
content = label_file.read().split("\n")
csvreader = reader(content, delimiter=",")
point_a = cast(Tuple[int, int], tuple(int(x) for x in next(csvreader)))
point_b = cast(Tuple[int, int], tuple(int(x) for x in next(csvreader)))
label_angle = radians(float(next(csvreader)[0]))
line = ImageLine.from_points(
image_shape=(obstacle_locator.image_width, obstacle_locator.image_height),
points=(point_a, point_b),
)
return line, label_angle
def test_horizon_angle(self):
"""Compares ``ObstacleLocator`` horizon estimates to ground truth annotations"""
uut_ol = ObstacleLocator(image_width=IMAGE_WIDTH, image_height=IMAGE_HEIGHT) # unit/module under test
for image_path, label_path in zip(DATASET_IMAGES_PATHS, DATASET_ANNOTATIONS_PATHS):
with self.subTest(image=image_path.name):
# Assert that we have the correct label for the test image
self.assertEqual(
image_path.name.split(".")[0],
label_path.name.split(".")[0],
msg="That isn't the right label for the image. This shouldn't happen.",
)
image = imread(image_path.as_posix())
# read annotation and test if ImageLine calculates the line's angle correctly
label_image_line, label_angle = self.parse_annotation(label_path.as_posix(), uut_ol)
self.assertAlmostEqual(label_angle, label_image_line.angle, places=2)
result = uut_ol.detect_horizon(image)
horizons = result[0]
# Test that a) a horizon is detected and b) it has the correct angle
self.assertTrue(len(horizons) > 0, msg="No horizon was detected.")
self.assertAlmostEqual(
horizons[0].angle, label_image_line.angle, places=1, msg="Horizon angle mismatch."
)
def test_missing_lines(self):
"""Tests the branch when no horizon line is detected in the image"""
uut_ol = ObstacleLocator(image_width=IMAGE_WIDTH, image_height=IMAGE_HEIGHT) # unit/module under test
image = imread(PATH_TO_FAILING.as_posix())
# ObstacleLocator does not find a horizon line in this image
result = uut_ol.detect_horizon(image)
self.assertFalse(result[0])