def part_4c():
"""Performs similar operations applied in part_4a using the images in DataSeq2.
This part implements the operations mentioned in the problem set which are similar to part_4a. In this case you
will use the images stored in the DataSeq2 directory and 0.png as the base image:
- 0.png
- 1.png
- 2.png
Make sure you explore different parameters and/or pre-process the input images to improve your results.
In this part you should save the following images:
- ps6-4-c-1.png
- ps6-4-c-2.png
Returns:
None
"""
dtsq2_01 = cv2.imread(os.path.join(input_dir, 'DataSeq2', '0.png'),
0) / 255.
dtsq2_02 = cv2.imread(os.path.join(input_dir, 'DataSeq2', '1.png'),
0) / 255.
dtsq2_03 = cv2.imread(os.path.join(input_dir, 'DataSeq2', '2.png'),
0) / 255.
levels = 5 # Define the number of levels
k_size = 15 # Select a kernel size
k_type = "uniform" # Select a kernel type
sigma = 0 # Select a sigma value if you are using a gaussian kernel
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
u1, v1 = ps6.hierarchical_lk(dtsq2_01, dtsq2_02, levels, k_size, k_type,
sigma, interpolation, border_mode)
jet_u1_v1 = jet_colormaps(u1, v1)
# You may want to try different parameters for the remaining function calls.
u2, v2 = ps6.hierarchical_lk(dtsq2_01, dtsq2_03, levels, k_size, k_type,
sigma, interpolation, border_mode)
jet_u2_v2 = jet_colormaps(u2, v2)
jets_stacked = np.concatenate((jet_u1_v1, jet_u2_v2), axis=0)
cv2.imwrite(os.path.join(output_dir, "ps6-4-c-1.png"), jets_stacked)
# Save difference between each warped image and original image (Shift0), stacked
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
shift_r2_warped = ps6.warp(dtsq2_02, u1, v1, interpolation, border_mode)
shift_r3_warped = ps6.warp(dtsq2_03, u2, v2, interpolation, border_mode)
diff_1_2 = shift_r2_warped - dtsq2_01
diff_1_3 = shift_r3_warped - dtsq2_01
diff_stacked = np.concatenate(
(ps6.normalize_and_scale(diff_1_2), ps6.normalize_and_scale(diff_1_3)),
axis=0)
cv2.imwrite(os.path.join(output_dir, "ps6-4-d-2.png"), diff_stacked)
def part_4a():
shift_0 = cv2.imread(os.path.join(input_dir, 'TestSeq', 'Shift0.png'),
0) / 255.
shift_r10 = cv2.imread(os.path.join(input_dir, 'TestSeq', 'ShiftR10.png'),
0) / 255.
shift_r20 = cv2.imread(os.path.join(input_dir, 'TestSeq', 'ShiftR20.png'),
0) / 255.
shift_r40 = cv2.imread(os.path.join(input_dir, 'TestSeq', 'ShiftR40.png'),
0) / 255.
levels = 4 # Define the number of levels
k_size = 25 # Select a kernel size
k_type = "uniform" # Select a kernel type
sigma = 0 # Select a sigma value if you are using a gaussian kernel
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
u10, v10 = ps6.hierarchical_lk(shift_0, shift_r10, levels, k_size, k_type,
sigma, interpolation, border_mode)
jet_u10_v10 = jet_colormaps(u10, v10)
# You may want to try different parameters for the remaining function calls.
u20, v20 = ps6.hierarchical_lk(shift_0, shift_r20, levels, k_size, k_type,
sigma, interpolation, border_mode)
jet_u20_v20 = jet_colormaps(u20, v20)
u40, v40 = ps6.hierarchical_lk(shift_0, shift_r40, levels, k_size, k_type,
sigma, interpolation, border_mode)
jet_u40_v40 = jet_colormaps(u40, v40)
jets_stacked = np.concatenate((jet_u10_v10, jet_u20_v20, jet_u40_v40),
axis=0)
cv2.imwrite(os.path.join(output_dir, "ps6-4-a-1.png"), jets_stacked)
# Save difference between each warped image and original image (Shift0), stacked
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
shift_r10_warped = ps6.warp(shift_r10, u10, v10, interpolation,
border_mode)
shift_r20_warped = ps6.warp(shift_r20, u20, v20, interpolation,
border_mode)
shift_r40_warped = ps6.warp(shift_r40, u40, v40, interpolation,
border_mode)
diff_0_10 = shift_r10_warped - shift_0
diff_0_20 = shift_r20_warped - shift_0
diff_0_40 = shift_r40_warped - shift_0
diff_stacked = np.concatenate((ps6.normalize_and_scale(diff_0_10),
ps6.normalize_and_scale(diff_0_20),
ps6.normalize_and_scale(diff_0_40)),
axis=0)
cv2.imwrite(os.path.join(output_dir, "ps6-4-a-2.png"), diff_stacked)
def part_5a():
"""Performs similar operations applied in part_4a using the images in Juggle.
This part implements the operations mentioned in the problem set which are similar to part_4a. In this case you
will use the images stored in the DataSeq2 directory and 0.png as the base image:
- 0.png
- 1.png
- 2.png
The sequence Juggle has a higher displacement between frames - the juggled balls move significantly. Try your
hierarchical LK on that sequence and see if you can warp frame 2 back to frame 1. Apply other techniques to make
your results better while using your LK methods.
Try to be creative on how you can achieve good results using optic flow.
In this part you should save the following images:
- ps6-5-a-1.png
- ps6-5-a-2.png
Returns:
None
"""
juggle_01 = cv2.imread(os.path.join(input_dir, 'Juggle', '0.png'),
0) / 255.
juggle_02 = cv2.imread(os.path.join(input_dir, 'Juggle', '1.png'),
0) / 255.
juggle_03 = cv2.imread(os.path.join(input_dir, 'Juggle', '2.png'),
0) / 255.
levels = 3 # Define the number of levels
k_size = 25 # Select a kernel size
k_type = "uniform" # Select a kernel type
sigma = 0 # Select a sigma value if you are using a gaussian kernel
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
u1, v1 = ps6.hierarchical_lk(juggle_01, juggle_02, levels, k_size, k_type,
sigma, interpolation, border_mode)
jet_u1_v1 = jet_colormaps(u1, v1)
# You may want to try different parameters for the remaining function calls.
u2, v2 = ps6.hierarchical_lk(juggle_01, juggle_03, levels, k_size, k_type,
sigma, interpolation, border_mode)
jet_u2_v2 = jet_colormaps(u2, v2)
jets_stacked = np.concatenate((jet_u1_v1, jet_u2_v2), axis=0)
cv2.imwrite(os.path.join(output_dir, "ps6-5-a-1.png"), jets_stacked)
# Save difference between each warped image and original image (Shift0), stacked
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
shift_r2_warped = ps6.warp(juggle_02, u1, v1, interpolation, border_mode)
shift_r3_warped = ps6.warp(juggle_03, u2, v2, interpolation, border_mode)
diff_1_2 = shift_r2_warped - juggle_01
diff_1_3 = shift_r3_warped - juggle_01
diff_stacked = np.concatenate(
(ps6.normalize_and_scale(diff_1_2), ps6.normalize_and_scale(diff_1_3)),
axis=0)
cv2.imwrite(os.path.join(output_dir, "ps6-5-a-2.png"), diff_stacked)
def part_3a_2():
"""Performs similar operations applied in part_3a_1 using the images DataSeq2/0, DataSeq2/1 and DataSeq2/2.
This part implements the operations mentioned in the problem set which are similar to part_3a_1. In this case you
will use the images stored in the DataSeq2 directory:
- 0.png
- 1.png
- 2.png
Make sure you explore different parameters and/or pre-process the input images to improve your results.
In this part you should save the following images:
- ps6-3-a-3.png
- ps6-3-a-4.png
Returns:
None
"""
dtsq2_01 = cv2.imread(os.path.join(input_dir, 'DataSeq2', '0.png'),
0) / 255.
dtsq2_02 = cv2.imread(os.path.join(input_dir, 'DataSeq2', '1.png'),
0) / 255.
dtsq2_03 = cv2.imread(os.path.join(input_dir, 'DataSeq2', '2.png'),
0) / 255.
# Todo: Your code here
levels = 4
dtsq2_01_g_pyr = ps6.gaussian_pyramid(dtsq2_01, levels)
dtsq2_02_g_pyr = ps6.gaussian_pyramid(dtsq2_02, levels)
dtsq2_03_g_pyr = ps6.gaussian_pyramid(dtsq2_03, levels)
level_id = 3 # Select the level number (or id) you wish to use
k_size = 15 # Select a kernel size
k_type = "uniform" # Select a kernel type
sigma = 0 # Select a sigma value if you are using a gaussian kernel
u, v = ps6.optic_flow_lk(dtsq2_01_g_pyr[level_id],
dtsq2_02_g_pyr[level_id], k_size, k_type,
sigma) # You may use different k_size and k_type
u, v = scale_u_and_v(u, v, level_id, dtsq2_02_g_pyr)
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
dtsq2_02_warped = ps6.warp(dtsq2_02, u, v, interpolation, border_mode)
scale = 3 # define a scale value
stride = 10 # define a stride value
dtsq2_01_02_flow = quiver(u, v, scale, stride)
diff_dtsq2_01_02 = dtsq2_01 - dtsq2_02_warped # difference image
# We will repeat the same process to obtain the difference image using dtsq2_02 and dtsq2_03
levels = 4 # Define the number of levels to build the gaussian pyramid
dtsq2_03_g_pyr = ps6.gaussian_pyramid(dtsq2_03, levels)
level_id = 3 # Select the level number (or id) you wish to use
k_size = 15 # Select a kernel size
k_type = "uniform" # Select a kernel type
sigma = 0 # Select a sigma value if you are using a gaussian kernel
u, v = ps6.optic_flow_lk(dtsq2_02_g_pyr[level_id],
dtsq2_03_g_pyr[level_id], k_size, k_type,
sigma) # You may use different k_size and k_type
u, v = scale_u_and_v(u, v, level_id, dtsq2_03_g_pyr)
scale = 3 # define a scale value
stride = 10 # define a stride value
dtsq2_02_03_flow = quiver(u, v, scale, stride)
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
dtsq2_03_warped = ps6.warp(dtsq2_03, u, v, interpolation, border_mode)
diff_dtsq2_02_03 = dtsq2_02 - dtsq2_03_warped
cv2.imwrite(os.path.join(output_dir, "ps6-3-a-3.png"),
np.concatenate((dtsq2_01_02_flow, dtsq2_02_03_flow), axis=0))
cv2.imwrite(
os.path.join(output_dir, "ps6-3-a-4.png"),
np.concatenate((ps6.normalize_and_scale(diff_dtsq2_01_02),
ps6.normalize_and_scale(diff_dtsq2_02_03)),
axis=0))
def part_3a_1():
yos_img_01, yos_img_01_g_pyr, yos_img_01_l_pyr = part_2a_2b(False)
yos_img_02 = cv2.imread(
os.path.join(input_dir, 'DataSeq1', 'yos_img_02.jpg'), 0) / 255.
levels = 4 # Define the number of levels to build the gaussian pyramid
yos_img_02_g_pyr = ps6.gaussian_pyramid(yos_img_02, levels)
level_id = 3 # Select the level number (or id) you wish to use
k_size = 15 # Select a kernel size
k_type = "uniform" # Select a kernel type
sigma = 0 # Select a sigma value if you are using a gaussian kernel
u, v = ps6.optic_flow_lk(yos_img_01_g_pyr[level_id],
yos_img_02_g_pyr[level_id], k_size, k_type,
sigma) # You may use different k_size and k_type
u, v = scale_u_and_v(u, v, level_id, yos_img_02_g_pyr)
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
yos_img_02_warped = ps6.warp(yos_img_02, u, v, interpolation, border_mode)
scale = 3 # define a scale value
stride = 10 # define a stride value
yos_img_01_02_flow = quiver(u, v, scale, stride)
diff_yos_img_01_02 = yos_img_01 - yos_img_02_warped # difference image
# We will repeat the same process to obtain the difference image using yos_img_02 and yos_img_03
yos_img_03 = cv2.imread(
os.path.join(input_dir, 'DataSeq1', 'yos_img_03.jpg'), 0) / 255.
levels = 4 # Define the number of levels to build the gaussian pyramid
yos_img_03_g_pyr = ps6.gaussian_pyramid(yos_img_03, levels)
level_id = 3 # Select the level number (or id) you wish to use
k_size = 15 # Select a kernel size
k_type = "uniform" # Select a kernel type
sigma = 0 # Select a sigma value if you are using a gaussian kernel
u, v = ps6.optic_flow_lk(yos_img_02_g_pyr[level_id],
yos_img_03_g_pyr[level_id], k_size, k_type,
sigma) # You may use different k_size and k_type
u, v = scale_u_and_v(u, v, level_id, yos_img_03_g_pyr)
scale = 3 # define a scale value
stride = 10 # define a stride value
yos_img_02_03_flow = quiver(u, v, scale, stride)
interpolation = cv2.INTER_CUBIC # Select an interpolation method (see cv2.remap)
border_mode = cv2.BORDER_REFLECT101 # Select a pixel extrapolation method (see cv2.remap)
yos_img_03_warped = ps6.warp(yos_img_03, u, v, interpolation, border_mode)
diff_yos_img_02_03 = yos_img_02 - yos_img_03_warped
cv2.imwrite(
os.path.join(output_dir, "ps6-3-a-1.png"),
np.concatenate((yos_img_01_02_flow, yos_img_02_03_flow), axis=0))
cv2.imwrite(
os.path.join(output_dir, "ps6-3-a-2.png"),
np.concatenate((ps6.normalize_and_scale(diff_yos_img_01_02),
ps6.normalize_and_scale(diff_yos_img_02_03)),
axis=0))