diff options
Diffstat (limited to 'test')
| -rw-r--r-- | test/context.py | 4 | ||||
| -rw-r--r-- | test/latent_space_test.py | 153 |
2 files changed, 143 insertions, 14 deletions
diff --git a/test/context.py b/test/context.py new file mode 100644 index 0000000..dd40aa2 --- /dev/null +++ b/test/context.py @@ -0,0 +1,4 @@ +import os +import sys +sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))) +import gantools diff --git a/test/latent_space_test.py b/test/latent_space_test.py index 0655923..d05eb78 100644 --- a/test/latent_space_test.py +++ b/test/latent_space_test.py @@ -1,9 +1,10 @@ import unittest import numpy as np from scipy.stats import truncnorm +import context from gantools import latent_space -from gantools import biggan import PIL.Image +import math def create_random_keyframe(n_vector, n_label): truncation = (0.9 - 0.1)*np.random.random() + 0.1 @@ -29,21 +30,145 @@ def save_ims(ims): i += 1 #### +def compare_float_arrays_2d(target_seq, actual_seq): + for target, actual in zip(target_seq, actual_seq): + for ti, ai in zip(target, actual): + assert math.isclose(ti, ai), 'target: %s; actual %s' % (str(target), str(actual)) + class TestLatentSpace(unittest.TestCase): - def test_sequence_keyframes(self): - num_frames = 20 - batch_size = 3 - keyframe_count = 3 - dim_z = 128 - dim_label = 1000 - keyframes = [create_random_keyframe(dim_z,dim_label) for i in range(keyframe_count)] - z_seq, label_seq, truncation_seq = latent_space.sequence_keyframes(keyframes, num_frames, batch_size) - self.assertIs(len(z_seq), num_frames) - self.assertIs(len(label_seq), num_frames) - gan = biggan.BigGAN() - ims = gan.sample(z_seq, label_seq, truncation_seq, batch_size) - save_ims(ims) + def test_linear_interp_basic(self): + target_seq = np.asarray([ + [0, 1, 2, 3, 4, 5, 6, 7, 8, 9], + [9, 8, 7, 6, 5, 4, 3, 2, 1, 0], + [1, 1, 1, 1, 1, 1, 1, 1, 1, 1], + ]).transpose() + points = np.asarray([target_seq[0], target_seq[-1]]) + step_count = target_seq.shape[0] + actual_seq = latent_space.linear_interp(points, step_count) + compare_float_arrays_2d(target_seq, actual_seq) + + def test_sequence_keyframes_linear_random_basic(self): + n_keyframes = 10 + n_vector = 100 + n_label = 1000 + num_frames = 100 + keyframes = np.asarray([create_random_keyframe(n_vector, n_label) for i in range(n_keyframes)]) + z, label, trunc = latent_space.sequence_keyframes( + keyframes, + num_frames, + batch_size=1, + interp_method='linear') + + assert (num_frames == z.shape[0]),\ + 'z sequence: target frame count: %s; actual shape: %s' % (num_frames, z.shape) + assert (num_frames == label.shape[0]),\ + 'label sequence: target frame count: %s; actual shape: %s' % (num_frames, label.shape) + assert (num_frames == trunc.shape[0]),\ + 'trunc sequence: target frame count: %s; actual shape: %s' % (num_frames, trunc.shape) + + def test_sequence_keyframes_linear_random_batch(self): + n_keyframes = 10 + n_vector = 100 + n_label = 1000 + num_frames = 100 + batch_size = 7 # pick something that doesn't divide num_frames + batch_div = int(num_frames // batch_size) + batch_rem = 1 if int(num_frames % batch_size) > 0 else 0 + batch_count = batch_div + batch_rem + keyframes = np.asarray([create_random_keyframe(n_vector, n_label) for i in range(n_keyframes)]) + z, label, trunc = latent_space.sequence_keyframes( + keyframes, + num_frames, + batch_size=batch_size, + interp_method='linear') + + assert (num_frames == z.shape[0]),\ + 'z sequence: target frame count: %s; actual shape: %s' % (num_frames, z.shape) + assert (num_frames == label.shape[0]),\ + 'label sequence: target frame count: %s; actual shape: %s' % (num_frames, label.shape) + assert (batch_count == trunc.shape[0]),\ + 'trunc sequence: target frame count: %s; actual shape: %s' % (batch_count, trunc.shape) + + def test_sequence_keyframes_linear_random_batch_oob(self): + n_keyframes = 10 + n_vector = 100 + n_label = 1000 + num_frames = 100 + batch_size = 150 # pick something that doesn't divide num_frames + batch_div = int(num_frames // batch_size) + batch_rem = 1 if int(num_frames % batch_size) > 0 else 0 + batch_count = batch_div + batch_rem + keyframes = np.asarray([create_random_keyframe(n_vector, n_label) for i in range(n_keyframes)]) + z, label, trunc = latent_space.sequence_keyframes( + keyframes, + num_frames, + batch_size=batch_size, + interp_method='linear') + + assert (num_frames == z.shape[0]),\ + 'z sequence: target frame count: %s; actual shape: %s' % (num_frames, z.shape) + assert (num_frames == label.shape[0]),\ + 'label sequence: target frame count: %s; actual shape: %s' % (num_frames, label.shape) + assert (batch_count == trunc.shape[0]),\ + 'trunc sequence: target frame count: %s; actual shape: %s' % (batch_count, trunc.shape) + + def test_sequence_keyframes_cubic_random_basic(self): + n_keyframes = 10 + n_vector = 100 + n_label = 1000 + num_frames = 100 + keyframes = np.asarray([create_random_keyframe(n_vector, n_label) for i in range(n_keyframes)]) + z, label, trunc = latent_space.sequence_keyframes( + keyframes, + num_frames, + batch_size=1, + interp_method='cubic') + + assert (num_frames == z.shape[0]),\ + 'z sequence: target frame count: %s; actual shape: %s' % (num_frames, z.shape) + assert (num_frames == label.shape[0]),\ + 'label sequence: target frame count: %s; actual shape: %s' % (num_frames, label.shape) + assert (num_frames == trunc.shape[0]),\ + 'trunc sequence: target frame count: %s; actual shape: %s' % (num_frames, trunc.shape) + + def test_sequence_keyframes_cubic_random_batch(self): + n_keyframes = 10 + n_vector = 100 + n_label = 1000 + num_frames = 100 + batch_size = 7 # pick something that doesn't divide num_frames + batch_div = int(num_frames // batch_size) + batch_rem = 1 if int(num_frames % batch_size) > 0 else 0 + batch_count = batch_div + batch_rem + keyframes = np.asarray([create_random_keyframe(n_vector, n_label) for i in range(n_keyframes)]) + z, label, trunc = latent_space.sequence_keyframes( + keyframes, + num_frames, + batch_size=batch_size, + interp_method='cubic') + + assert (num_frames == z.shape[0]),\ + 'z sequence: target frame count: %s; actual shape: %s' % (num_frames, z.shape) + assert (num_frames == label.shape[0]),\ + 'label sequence: target frame count: %s; actual shape: %s' % (num_frames, label.shape) + assert (batch_count == trunc.shape[0]),\ + 'trunc sequence: target frame count: %s; actual shape: %s' % (batch_count, trunc.shape) + def test_sequence_keyframes_cubic_random_batch_oob(self): + n_keyframes = 10 + n_vector = 100 + n_label = 1000 + num_frames = 100 + batch_size = 150 # pick something that doesn't divide num_frames + batch_div = int(num_frames // batch_size) + batch_rem = 1 if int(num_frames % batch_size) > 0 else 0 + batch_count = batch_div + batch_rem + keyframes = np.asarray([create_random_keyframe(n_vector, n_label) for i in range(n_keyframes)]) + z, label, trunc = latent_space.sequence_keyframes( + keyframes, + num_frames, + batch_size=batch_size, + interp_method='cubic') if __name__ == '__main__': |