def run(src_path, label_path, sigmas, epsilon=0.02, same=False):
src_name = os.path.basename(src_path)
src_name = os.path.splitext(src_name)[0]
label_name = os.path.basename(label_path)
label_name = os.path.splitext(label_name)[0]
src_C_8U = loadRGB(src_path)
src_C_32F = to32F(src_C_8U)
label_C_8U = loadRGB(label_path)
label_C_32F = to32F(label_C_8U)
os.makedirs('./output/', exist_ok=True)
for sigma in sigmas:
print(sigma)
for eps in epsilon:
if same:
guided_filter = GuidedFilter(label_C_32F,
radius=sigma,
epsilon=eps)
else:
guided_filter = GuidedFilter(src_C_32F,
radius=sigma,
epsilon=eps)
result = guided_filter.filter(label_C_32F)
cv2.imwrite(
'./output/{}_radius{}_eps{:.03f}.png'.format(
label_name, sigma, eps),
cv2.cvtColor(result * 255, cv2.COLOR_RGB2GRAY))
def GuidedFilter(input_imagefile, guidance_imgfile):
image_name = os.path.basename(input_imagefile)
image_name = os.path.splitext(image_name)[0]
C_Input_8U = loadRGB(input_imagefile)
C_Input_32F = to32F(C_Input_8U)
C_Guidnace_8U = loadRGB(guidance_imgfile)
C_Guidnace_32F = to32F(C_Guidnace_8U)
aspect = C_Input_32F.shape[0] / float(C_Input_32F.shape[1])
fig_width = 10
fig_height = int(2 * fig_width * aspect / 3) + 2
fig = plt.figure(figsize=(fig_width, fig_height))
fig.subplots_adjust(left=0.05,
bottom=0.05,
right=0.95,
top=0.82,
wspace=0.02,
hspace=0.3)
h, w = C_Input_32F.shape[:2]
image_size_str = "Image size: %s x %s" % (w, h)
fig.suptitle("Filtering noise image\n%s" % image_size_str)
# plt.subplot(231)
# plt.title("Original")
# plt.imshow(C_Input_32F)
# plt.axis('off')
h, w, cs = C_Input_32F.shape
plt.subplot(232)
plt.title("input img")
plt.imshow(C_Input_32F)
plt.axis('off')
sigmas = [5, 10, 20]
plot_id = 234
for sigma in sigmas:
guided_filter = FastGuidedFilter(C_Guidnace_32F,
radius=sigma,
epsilon=0.02)
C_smooth = guided_filter.filter(C_Input_32F)
C_smooth = np.clip(C_smooth, 0.0, 1.0)
plt.subplot(plot_id)
plt.title("Filtered ($r$=%s)" % sigma)
plt.imshow(C_smooth)
plt.axis('off')
plot_id += 1
result_file = resultFile(image_name)
plt.savefig(result_file)
def loadData(data_name, data_id):
data_file = dataFile(data_name, data_id)
if data_file is None:
return None
return loadRGB(data_file)
def postResize(self):
print(" Post resize")
data_name = self._keyword
data_files = dataFiles(data_name)
for data_file in data_files:
data_filename = os.path.basename(data_file)
C_8U = loadRGB(data_file)
if C_8U is None:
os.remove(data_file)
print(" - Delete: {}").format(data_filename)
continue
h, w = C_8U.shape[0:2]
opt_scale = 800.0 / float(h)
opt_scale = max(opt_scale, 800.0 / float(w))
opt_scale = min(opt_scale, 1.0)
h_opt = int(opt_scale * h)
w_opt = int(opt_scale * w)
C_8U_small = cv2.resize(C_8U, (w_opt, h_opt))
saveRGB(data_file, C_8U_small)
print(" - Resized: {}").format(data_filename)
def loadData(data_name, data_id):
data_file = dataFile(data_name, data_id)
if data_file is None:
return None
return loadRGB(data_file)
def postResize(self):
print " Post resize"
data_name = self._keyword
data_files = dataFiles(data_name)
for data_file in data_files:
data_filename = os.path.basename(data_file)
C_8U = loadRGB(data_file)
if C_8U is None:
os.remove(data_file)
print " - Delete: %s" % data_filename
continue
h, w = C_8U.shape[0:2]
opt_scale = 800.0 / float(h)
opt_scale = max(opt_scale, 800.0 / float(w))
opt_scale = min(opt_scale, 1.0)
h_opt = int(opt_scale * h)
w_opt = int(opt_scale * w)
C_8U_small = cv2.resize(C_8U, (w_opt, h_opt))
saveRGB(data_file, C_8U_small)
print " - Resized: %s" % data_filename
def performanceTest(image_file):
C_8U = loadRGB(image_file)
C_32F = to32F(C_8U)
h, w = C_32F.shape[:2]
image_size_str = "Image size: %s x %s" %(w, h)
fig, axes = plt.subplots(nrows=2, ncols=1, figsize=(8, 8))
fig.subplots_adjust(left=0.1, right=0.7, top=0.86, hspace=0.4)
fig.suptitle("Peformance of guided filter\n%s" % image_size_str)
filter_types = {"Bilateral Filter": (BilateralFilter, "r"),
"Guided Filter": (GuidedFilter, "g"),
"Fast Guided Filter": (FastGuidedFilter, "b")}
sigmas = range(3, 31, 2)
axes[0].set_title('For small radius $r$')
performanceTestSigmas(C_32F, sigmas, filter_types, axes[0])
sigmas = range(10, 100, 5)
filter_types = {"Guided Filter": (GuidedFilter, "g"),
"Fast Guided Filter": (FastGuidedFilter, "b")}
axes[1].set_title('For large radius $r$')
performanceTestSigmas(C_32F, sigmas, filter_types, axes[1])
result_name = "performance"
result_file = resultFile(result_name)
plt.savefig(result_file)
def runSmoothNoiseResult(image_file):
image_name = os.path.basename(image_file)
image_name = os.path.splitext(image_name)[0]
C_8U = loadRGB(image_file)
C_32F = to32F(C_8U)
aspect = C_32F.shape[0] / float(C_32F.shape[1])
fig_width = 10
fig_height = int(2 * fig_width * aspect / 3) + 2
fig = plt.figure(figsize=(fig_width, fig_height))
fig.subplots_adjust(left=0.05, bottom=0.05, right=0.95, top=0.82, wspace=0.02, hspace=0.3)
h, w = C_32F.shape[:2]
image_size_str = "Image size: %s x %s" %(w, h)
fig.suptitle("Filtering noise image\n%s" % image_size_str)
plt.subplot(231)
plt.title("Original")
plt.imshow(C_32F)
plt.axis('off')
h, w, cs = C_32F.shape
C_noise = np.float32(C_32F + 0.3 * np.random.rand(h, w, cs))
C_noise = np.clip(C_noise, 0.0, 1.0)
plt.subplot(232)
plt.title("Noise")
plt.imshow(C_noise)
plt.axis('off')
sigmas = [5, 10, 20]
plot_id = 234
for sigma in sigmas:
guided_filter = FastGuidedFilter(C_noise, radius=sigma, epsilon=0.02)
C_smooth = guided_filter.filter(C_noise)
C_smooth = np.clip(C_smooth, 0.0, 1.0)
plt.subplot(plot_id)
plt.title("Filtered ($r$=%s)" %sigma)
plt.imshow(C_smooth)
plt.axis('off')
plot_id +=1
result_file = resultFile(image_name)
plt.savefig(result_file)
def performanceTest(image_file):
C_8U = loadRGB(image_file)
C_32F = to32F(C_8U)
h, w = C_32F.shape[:2]
image_size_str = "Image size: %s x %s" % (w, h)
fig, axes = plt.subplots(nrows=2, ncols=1, figsize=(8, 8))
fig.subplots_adjust(left=0.1, right=0.7, top=0.86, hspace=0.4)
fig.suptitle("Peformance of guided filter\n%s" % image_size_str)
filter_types = {
"Bilateral Filter": (BilateralFilter, "r"),
"Guided Filter": (GuidedFilter, "g"),
"Fast Guided Filter": (FastGuidedFilter, "b")
}
sigmas = range(3, 31, 2)
axes[0].set_title('For small radius $r$')
performanceTestSigmas(C_32F, sigmas, filter_types, axes[0])
sigmas = range(10, 100, 5)
filter_types = {
"Guided Filter": (GuidedFilter, "g"),
"Fast Guided Filter": (FastGuidedFilter, "b")
}
axes[1].set_title('For large radius $r$')
performanceTestSigmas(C_32F, sigmas, filter_types, axes[1])
result_name = "performance"
result_file = resultFile(result_name)
plt.savefig(result_file)
