def test_expected_time(self):
d = PerfectDetector(self.train_dataset, self.train_dataset, 'dog')
img = Image(500, 375, self.train_dataset.classes, 'test')
print d.expected_time(img)
assert(d.expected_time(img) == 10)
img = Image(250, 375, self.train_dataset.classes, 'test')
assert(d.expected_time(img) == 5)
def test_get_random_windows(self):
image = Image(width=3, height=2, classes=[], name='test')
window_params = WindowParams(
min_width=2, stride=1, scales=[1, 0.5], aspect_ratios=[0.5])
windows = image.get_random_windows(window_params, 2)
assert(windows.shape[0] == 2)
windows = image.get_random_windows(window_params, 3)
assert(windows.shape[0] == 3)
def test_get_det_gt(self):
image = Image.load_from_json(self.classes, self.data)
objects_table = Table(np.array([
[0, 0, 0, 0, 0, 0, 0],
[1, 1, 1, 1, 1, 0, 0],
[2, 2, 2, 2, 2, 0, 0]]), self.columns)
assert(image.get_objects() == objects_table)
data = self.data.copy()
data['objects'][0]['diff'] = 1
data['objects'][1]['trun'] = 1
image = Image.load_from_json(self.classes, data)
objects_table = Table(np.array([
[0, 0, 0, 0, 0, 1, 0],
[1, 1, 1, 1, 1, 0, 1],
[2, 2, 2, 2, 2, 0, 0]]), self.columns)
assert(image.get_objects(
with_diff=True, with_trun=True) == objects_table)
objects_table = Table(np.array([
[1, 1, 1, 1, 1, 0, 1],
[2, 2, 2, 2, 2, 0, 0]]), self.columns)
assert(image.get_objects(
with_diff=False, with_trun=True) == objects_table)
# this should be default behavior
assert(image.get_objects() == objects_table)
objects_table = Table(np.array([
[2, 2, 2, 2, 2, 0, 0]]), self.columns)
assert(image.get_objects(
with_diff=False, with_trun=False) == objects_table)
objects_table = Table(np.array([
[0, 0, 0, 0, 0, 1, 0],
[2, 2, 2, 2, 2, 0, 0]]), self.columns)
assert(image.get_objects(
with_diff=True, with_trun=False) == objects_table)
# What if everything is filtered out?
data['objects'] = data['objects'][:-1]
objects_table = Table(np.array([
[0, 0, 0, 0, 0, 1, 0],
[1, 1, 1, 1, 1, 0, 1]]), self.columns)
image = Image.load_from_json(self.classes, data)
assert(image.get_objects(
with_diff=True, with_trun=True) == objects_table)
assert(
image.get_objects(with_diff=False, with_trun=False).shape[0] == 0)
def test_get_windows_lots(self):
t = time.time()
image = Image(width=640, height=480, classes=[], name='test')
window_params = WindowParams()
window_params.min_width = 10
window_params.stride = 8
window_params.aspect_ratios = [0.5, 1, 1.5]
window_params.scales = 1. / 2 ** np.array([0, 0.5, 1, 1.5, 2])
print(window_params)
windows = image.get_windows(window_params)
time_passed = time.time() - t
print("Generating windows took %.3f seconds" % time_passed)
print(np.shape(windows))
print(windows[:10, :])
rand_ind = np.random.permutation(np.shape(windows)[0])[:10]
print(windows[rand_ind, :])
def load_from_pascal(self, name, force=False):
"""
Load PASCAL VOC data, loading from cache if it exists at a canonical path.
Both the PASCAL and cache paths should be set in self.config.
After loading, saves the images and classes to cache.
Uses Image class to load data from the PASCAL XML files.
Args:
name (string): in ['train','trainval','val','test']
force (bool): [optional] if True, do not load from cache
Returns:
self
Raises:
self
"""
tt = TicToc().tic()
print("Dataset: %s" % self.get_name())
filename = self.config.get_cached_dataset_filename(self.name)
if os.path.exists(filename) and not force:
with open(filename) as f:
cached = cPickle.load(f)
self.classes = cached['classes']
self.images = cached['images']
print("...loaded from cache in %.2f s" % tt.qtoc())
else:
print("...loading from scratch")
self.classes = self.config.pascal_classes
with open(self.config.get_pascal_path(name)) as f:
imgset = [line.strip() for line in f.readlines() if len(line) > 0]
images_dir = os.path.join(self.config.pascal_dir, "JPEGImages")
for i, img in enumerate(imgset):
tt.running(
'loading', "...on image {0}/{1}".format(i, len(imgset)))
xml_filename = os.path.join(
self.config.pascal_dir, 'Annotations', img + '.xml')
self.images.append(
Image.load_from_pascal_xml_filename(self.classes, xml_filename, images_dir))
print("...saving to cache file")
data = {'classes': self.classes, 'images': self.images}
filename = self.config.get_cached_dataset_filename(self.name)
with open(filename, 'w') as f:
cPickle.dump(data, f)
print("...done in %.2f s\n" % tt.qtoc())
return self.after_load()
def test_load_json_data(self):
image = Image.load_from_json(self.classes, self.data)
assert(image.width == 640 and image.height == 480)
assert(image.classes == ['A', 'B', 'C'])
assert(image.name == 'test_image')
objects_table = Table(np.array([
[0, 0, 0, 0, 0, 0, 0],
[1, 1, 1, 1, 1, 0, 0],
[2, 2, 2, 2, 2, 0, 0]]), self.columns)
assert(image.objects_table == objects_table)
def test_get_windows(self):
image = Image(3,2,['whatever'],'test')
window_params = WindowParams(
min_width=2,stride=1,scales=[1,0.5],aspect_ratios=[0.5])
windows = image.get_windows(window_params)
correct = np.array(
[[ 0., 0., 2., 2.],
[ 0., 0., 3., 2.],
[ 1., 0., 3., 2.],
[ 2., 0., 2., 2.],
[ 0., 1., 2., 2.],
[ 0., 1., 3., 2.],
[ 1., 1., 3., 2.],
[ 2., 1., 2., 2.],
[ 0., 0., 3., 3.],
[ 0., 0., 4., 3.]])
assert(windows.shape[0] > 0)
print(windows)
assert(np.all(correct == windows))
def test_get_cls_counts_and_gt(self):
data = self.data.copy()
image = Image.load_from_json(self.classes, data)
assert(np.all(image.get_cls_counts() == np.array([1, 1, 1])))
assert(np.all(image.get_cls_gt() == np.array([True, True, True])))
assert(image.contains_class('A'))
assert(image.contains_class('B'))
data['objects'][0]['class'] = 'B'
image = Image.load_from_json(self.classes, data)
# doesn't actually have to be Series, can be ndarray for comparison
assert(np.all(image.get_cls_counts() == np.array([0, 2, 1])))
assert(np.all(image.get_cls_gt() == np.array([False, True, True])))
assert(not image.contains_class('A'))
assert(image.contains_class('B'))
data['objects'] = []
image = Image.load_from_json(self.classes, data)
assert(np.all(image.get_cls_counts() == np.array([0, 0, 0])))
assert(np.all(image.get_cls_gt() == np.array([False, False, False])))
assert(not image.contains_class('A'))
assert(not image.contains_class('B'))
def generate_synthetic(self):
"""
Generate a synthetic dataset of 4 classes that follows some simple
but strong cooccurence rules.
Returns
-------
self
"""
num_images = 1000
choice_probs = {
(1, 0, 0, 0): 2,
(0, 1, 0, 0): 1,
(0, 0, 1, 0): 1,
(0, 0, 0, 1): 1,
(1, 1, 0, 0): 2,
(1, 0, 1, 0): 0,
(1, 0, 0, 1): 0,
(0, 1, 1, 0): 0,
(0, 1, 0, 1): 1,
(0, 0, 1, 1): 2,
(1, 1, 1, 0): 0,
(1, 1, 0, 1): 1,
(1, 0, 1, 1): 1,
(0, 1, 1, 1): 2}
probs = np.array(choice_probs.values())
cum_probs = np.cumsum(1. * probs / np.sum(probs))
self.classes = ['A', 'B', 'C', 'D']
for i in range(0, num_images):
image = Image(100, 100, self.classes, str(i))
choice = np.where(cum_probs > np.random.rand())[0][0]
objects = []
for cls_ind, clas in enumerate(choice_probs.keys()[choice]):
if clas == 1:
objects.append(np.array([0, 0, 0, 0, cls_ind, 0, 0]))
image.objects_table = Table(np.array(objects), Image.columns)
self.images.append(image)
return self.after_load()
def load_from_json(self, filename):
"""
Load all parameters of the dataset from a JSON file.
Parameters
----------
filename: string
path to JSON data file
Returns
-------
self
"""
with open(filename) as f:
json_data = json.load(f)
self.classes = json_data['classes']
self.images = [Image.load_from_json(
self.classes, image) for image in json_data['images']]
return self.after_load()
def test_get_whole_image_bbox(self):
image = Image(20, 10, [], 'test_image')
assert(image.get_whole_image_bbox() == BoundingBox((0, 0, 20, 10)))
image = Image(2, 100, [], 'test_image')
assert(image.get_whole_image_bbox() == BoundingBox((0, 0, 2, 100)))
