def pyrdown_hybrid_test():
image = imread('test_materials/MonroeEnstein_AudeOliva2007.jpg')
image = pyrdown(image)
image = pyrdown(image)
image = pyrdown(image)
image = np.uint8(np.clip(image, 0, 255))
correct = imread('test_materials/MonroeEnstein_AudeOliva2007_small.png')
assert (np.abs(image - correct) / 255.0 < 1e-2).all()
def pyrdown_hybrid_test():
image = imread('test_materials/MonroeEnstein_AudeOliva2007.jpg')
image = image.astype(np.float32)
image = pyrdown(image)
image = pyrdown(image)
image = pyrdown(image)
image = np.uint8(np.clip(image, 0, 255))
correct = imread('test_materials/MonroeEnstein_AudeOliva2007_small.png')
assertNear(image / 255.0, correct / 255.0, 1e-2)
def pyrdown_hybrid_test():
image = imread('test_materials/MonroeEnstein_AudeOliva2007.jpg')
image = image.astype(np.float32)
image = pyrdown(image)
image = pyrdown(image)
image = pyrdown(image)
image = np.uint8(np.clip(image, 0, 255))
correct = imread('test_materials/MonroeEnstein_AudeOliva2007_small.png')
assertNear(image / 255.0, correct / 255.0, 1e-2)
def pyrdown_random_test():
height = 16
width = 31
image = np.random.random((height, width, 3))
down = pyrdown(image)
assert down.shape == ((height + 1) / 2, (width + 1) / 2, 3)
def pyrdown_random_test():
height = 16
width = 31
image = np.random.random((height, width, 3))
down = pyrdown(image)
assert down.shape == ((height + 1) / 2, (width + 1) / 2, 3)
def pyrdown_odd_test():
height = 17
width = 17
image = np.ones((height, width, 3), dtype=np.float32)
down = pyrdown(image)
assert down.shape == ((height + 1) / 2, (width + 1) / 2, 3)
assert (down == 1).all()
def pyrdown_nonsquare_test():
height = 16
width = 31
image = np.ones((height, width, 3), dtype=np.float32)
down = pyrdown(image)
assert down.shape == ((height + 1) // 2, (width + 1) // 2, 3)
assert (down == 1).all()
def pyrdown_even_test():
height = 16
width = 16
image = np.ones((height, width, 3), dtype=np.float32)
down = pyrdown(image)
assert down.shape == (height / 2, width / 2, 3)
assertNear(down, 1, 1e-6)
def pyrdown_even_test():
height = 16
width = 16
image = np.ones((height, width, 3), dtype=np.float32)
down = pyrdown(image)
assert down.shape == (height / 2, width / 2, 3)
assert (down == 1).all()
def pyrdown_nonsquare_test():
height = 16
width = 31
image = np.ones((height, width, 3), dtype=np.float32)
down = pyrdown(image)
assert down.shape == ((height + 1) / 2, (width + 1) / 2, 3)
assertNear(down, 1, 1e-6)
def pyrdown_pyrup_test_single_channel():
height = 16
width = 31
image = np.random.random((height * 2, width * 2, 1))
down = pyrdown(image)
assert down.shape == (height, width, 1)
up = pyrup(down)
assert up.shape == (height * 2, width * 2, 1)
def pyrup_pyrdown_test_single_channel():
height = 16
width = 31
image = np.random.random((height, width, 1))
up = pyrup(image)
assert up.shape == (2 * height, 2 * width, 1)
down = pyrdown(up)
assert down.shape == (height, width, 1)
def pyrup_pyrdown_test():
height = 16
width = 31
image = np.random.random((height, width, 3))
up = pyrup(image)
assert up.shape == (2 * height, 2 * width, 3)
down = pyrdown(up)
assert down.shape == (height, width, 3)
albedo = imread(data.albedo_png)
normals = np.load(data.normals_npy)
if mode in ('depth', 'both'):
depth = np.load(data.depth_npy)
K_right = data.K_right
depth = cv2.medianBlur(depth, 5)
depth = cv2.medianBlur(depth, 5)
if data.mesh_downscale_factor > data.stereo_downscale_factor:
for i in range(data.mesh_downscale_factor -
data.stereo_downscale_factor):
# depth is 2D so give it 1 channel and then extract it back out
x = depth[:, :, np.newaxis]
y = pyrdown(x)
depth = y[:, :, 0]
elif data.stereo_downscale_factor > data.mesh_downscale_factor:
for i in range(data.stereo_downscale_factor -
data.mesh_downscale_factor):
x = depth[:, :, np.newaxis]
y = pyrup(x)
depth = y[:, :, 0]
for i in range(data.mesh_downscale_factor):
K_right[:2, :] /= 2
if mode == 'both':
depth_weight = data.depth_weight
if mode == 'depth':
Rt_left = data.Rt_left
ncc_size = data.ncc_size
width = data.width
height = data.height
ncc_gif_writer = GifWriter(data.ncc_temp, data.ncc_gif)
projected_gif_writer = GifWriter(data.projected_temp, data.projected_gif)
"""
Images are pretty large, so we're going to reduce their size
in each dimension.
"""
right = data.right[0]
left = data.left[0]
for i in xrange(data.stereo_downscale_factor):
right = pyrdown(right)
left = pyrdown(left)
height, width, _ = right.shape
K_left[:2, :] /= 2
K_right[:2, :] /= 2
"""
We'll give you the depth labels for this problem.
"""
depths = get_depths(data)
tic = time.time()
"""
The planes will be swept fronto-parallel to the right camera, so no
reprojection needs to be done for this image. Simply compute the normalized
patches across the entire image.
"""
cam0 = calib_values['cam0']
cam1 = calib_values['cam1']
doffs = float(calib_values['doffs'])
baseline = float(calib_values['baseline'])
width = int(calib_values['width'])
height = int(calib_values['height'])
# since this is a String, work with it to obtain f value
cam0 = cam0[1:-1]
tokens = cam0.split(' ')
f = float(tokens[0])
# reduce size of ground truth image using the pyrdown from util.py
# reduce size by same amount as the plane_sweep_stereo did
for i in range(data.stereo_downscale_factor):
ground_truth = pyrdown(ground_truth)
# we have some infinite values so set them to the max value we find in the ground truth
ground_truth[ground_truth == float('inf')] = 0
ground_truth[ground_truth == 0] = np.max(ground_truth)
# convert our depths to disparity
our_depth = (baseline * f) / our_depth
print('done converting disparity to depth')
# compute root mean squared error
N = np.size(ground_truth)
h, w = ground_truth.shape
sum = 0
for y in range(h):
tic = time.time()
if mode in ('normals', 'both'):
albedo = imread(data.albedo_png)
normals = np.load(data.normals_npy)
if mode in ('depth', 'both'):
depth = np.load(data.depth_npy)
K_right = data.K_right
depth = cv2.medianBlur(depth, 5)
depth = cv2.medianBlur(depth, 5)
if data.mesh_downscale_factor > data.stereo_downscale_factor:
for i in xrange(data.mesh_downscale_factor -
data.stereo_downscale_factor):
depth = pyrdown(depth)
elif data.stereo_downscale_factor > data.mesh_downscale_factor:
for i in xrange(data.stereo_downscale_factor -
data.mesh_downscale_factor):
depth = pyrup(depth)
for i in xrange(data.mesh_downscale_factor):
K_right[:2, :] /= 2
if mode == 'both':
depth_weight = data.depth_weight
if mode == 'depth':
albedo = data.right[0]
if alpha is not None:
tic = time.time()
if mode in ('normals', 'both'):
albedo = imread(data.albedo_png)
normals = np.load(data.normals_npy)
if mode in ('depth', 'both'):
depth = np.load(data.depth_npy)
K_right = data.K_right
depth = cv2.medianBlur(depth, 5)
depth = cv2.medianBlur(depth, 5)
if data.mesh_downscale_factor > data.stereo_downscale_factor:
for i in range(data.mesh_downscale_factor -
data.stereo_downscale_factor):
depth = pyrdown(depth)
elif data.stereo_downscale_factor > data.mesh_downscale_factor:
for i in range(data.stereo_downscale_factor -
data.mesh_downscale_factor):
depth = pyrup(depth)
for i in range(data.mesh_downscale_factor):
K_right[:2, :] /= 2
if mode == 'both':
depth_weight = data.depth_weight
if mode == 'depth':
albedo = data.right[0]
if alpha is not None:
ncc_size = data.ncc_size
width = data.width
height = data.height
ncc_gif_writer = GifWriter(data.ncc_temp, data.ncc_gif)
projected_gif_writer = GifWriter(data.projected_temp, data.projected_gif)
"""
Images are pretty large, so we're going to reduce their size
in each dimension.
"""
right = data.right[0]
left = data.left[0]
for i in xrange(data.stereo_downscale_factor):
right = pyrdown(right)
left = pyrdown(left)
height, width, _ = right.shape
K_left[:2, :] /= 2
K_right[:2, :] /= 2
"""
We'll give you the depth labels for this problem.
"""
depths = get_depths(data)
tic = time.time()
"""
The planes will be swept fronto-parallel to the right camera, so no
reprojection needs to be done for this image. Simply compute the normalized
