''' Runs unit tests for various SPy spatial functions. To run the unit tests, type the following from the system command line: # python -m spectral.tests.spatial ''' from __future__ import absolute_import, division, print_function, unicode_literals import numpy as np from numpy.testing import assert_allclose from warnings import warn import spectral as spy from spectral.algorithms.spatial import (get_window_bounds, get_window_bounds_clipped, map_class_ids, map_classes, map_window) from spectral.tests.spytest import SpyTest class SpatialWindowTest(SpyTest): '''Tests various spatial functions.''' def setup(self): self.data = spy.open_image('92AV3C.lan').load() def test_get_window_bounds(self): assert(get_window_bounds(90, 90, 3, 7, 30, 40) == (29, 32, 37, 44)) def test_get_window_bounds_border(self): assert(get_window_bounds(90, 90, 3, 7, 0, 2) == (0, 3, 0, 7)) def test_get_window_bounds_clipped(self): assert(get_window_bounds_clipped(90, 90, 3, 7, 30, 40) \ == (29, 32, 37, 44)) def test_get_window_bounds_clipped_border(self): assert(get_window_bounds_clipped(90, 90, 3, 7, 0, 2) == (0, 2, 0, 6)) def test_map_window(self): '''Test computing spectra average over local window.''' f = lambda X, ij: np.mean(X.reshape((-1, X.shape[-1])), axis=0) X = self.data y = map_window(f, X, (3, 5), (10, 50), (20, 40)) t = np.mean(X[9:12, 18:23].reshape((-1, X.shape[-1])), axis=0) assert_allclose(y[0, 0], t) def test_map_window_clipped(self): '''Test spatial averaging near border with clipped window.''' f = lambda X, ij: np.mean(X.reshape((-1, X.shape[-1])), axis=0) X = self.data y = map_window(f, X, (3, 5), (100, None), (100, None), border='clip') t = np.mean(X[-2:, -3:].reshape((-1, X.shape[-1])), axis=0) assert_allclose(y[-1, -1], t) def test_map_window_shifted(self): '''Test spatial averaging near border with shifted window.''' f = lambda X, ij: np.mean(X.reshape((-1, X.shape[-1])), axis=0) X = self.data y = map_window(f, X, (3, 5), (100, None), (100, None), border='shift') t = np.mean(X[-3:, -5:].reshape((-1, X.shape[-1])), axis=0) assert_allclose(y[-1, -1], t) def test_map_window_stepped(self): '''Test spatial averaging with non-unity row/column step sizes.''' f = lambda X, ij: np.mean(X.reshape((-1, X.shape[-1])), axis=0) X = self.data y = map_window(f, X, (3, 5), (30, 60, 3), (70, 100, 4), border='shift') t = np.mean(X[32:35, 72:77].reshape((-1, X.shape[-1])), axis=0) assert_allclose(y[1, 1], t) class MapClassesTest(SpyTest): '''Test mapping of class indices between classification images.''' def setup(self): self.gt = spy.open_image('92AV3GT.GIS').read_band(0) def test_map_class_ids_identity(self): '''Mapping a class image back to itself should yield identity map.''' gt = np.array(self.gt) d = map_class_ids(gt, gt) for i in set(gt.ravel()): assert(i in d) for (i, j) in d.items(): assert(j == i) def test_map_class_ids_identity_unlabeled(self): '''Mapping a class image back to itself with an unlabeled class.''' gt = np.array(self.gt) d = map_class_ids(gt, gt, unlabeled=0) for i in set(gt.ravel()): assert(i in d) for (i, j) in d.items(): assert(j == i) def test_map_class_ids_identity_multiple_unlabeled(self): '''Mapping a class image back to itself with unlabeled classes.''' gt = np.array(self.gt) d = map_class_ids(gt, gt, unlabeled=[2, 4]) for i in set(gt.ravel()): assert(i in d) for (i, j) in d.items(): assert(j == i) def test_map_class_ids_isomorphic(self): '''Test map_class_ids with isomorphic classes.''' gt = np.array(self.gt) gt2 = gt + 1 d = map_class_ids(gt, gt2) for (i, j) in d.items(): assert(j == i + 1) def test_map_class_ids_isomorphic_background(self): '''Test map_class_ids with isomorphic classes and background arg.''' gt = np.array(self.gt) gt2 = gt + 1 d = map_class_ids(gt, gt2, unlabeled=0) assert(d[0] == 0) d.pop(0) for (i, j) in d.items(): assert(j == i + 1) def test_map_class_ids_src_gt_dest(self): '''Test map_class_ids with more classes in source image.''' gt = np.array(self.gt) (i, j) = (100, 30) old_label = gt[i, j] new_label = max(set(gt.ravel())) + 10 gt2 = np.array(gt) gt2[i, j] = new_label d = map_class_ids(gt2, gt) # There are enough pixels for each class that a new single-pixel class # should not be mapped to one of the existing classes. assert(d[new_label] not in gt) d.pop(new_label) for (i, j) in d.items(): assert(j == i) def test_map_class_ids_dest_gt_src(self): '''Test map_class_ids with more classes in dest image.''' gt = np.array(self.gt) (i, j) = (100, 30) old_label = gt[i, j] new_label = max(set(gt.ravel())) + 10 gt2 = np.array(gt) gt2[i, j] = new_label d = map_class_ids(gt, gt2) for (i, j) in d.items(): assert(j == i) def test_map_classes_isomorphic(self): '''map_classes should map isomorphic class image back to original.''' gt = np.array(self.gt) gt2 = gt + 1 d = map_class_ids(gt2, gt) result = map_classes(gt2, d) assert(np.alltrue(result == gt)) def test_map_fails_allow_unmapped_false(self): '''map_classes should raise ValueError if image has unmapped value.''' gt = np.array(self.gt) gt2 = gt + 1 d = map_class_ids(gt2, gt) d.pop(1) try: result = map_classes(gt2, d) except ValueError: pass else: assert(False) def test_map_allow_unmapped_true(self): '''map_classes should raise ValueError if image has unmapped value.''' gt = np.array(self.gt) gt2 = gt + 1 d = map_class_ids(gt2, gt) d.pop(1) result = map_classes(gt2, d, allow_unmapped=True) assert(np.alltrue(result[gt2 == 1] == 1)) def run(): print('\n' + '-' * 72) print('Running spatial tests.') print('-' * 72) for T in [SpatialWindowTest, MapClassesTest]: T().run() if __name__ == '__main__': from spectral.tests.run import parse_args, reset_stats, print_summary parse_args() reset_stats() run() print_summary()