def visualize_points_images():
features, labels = get_dataset(1000)
for index, (f, l) in enumerate(zip(features, labels)):
if l.classification == 0:
continue
f = visualize_points_one_image(f, l, index)
cv2.imwrite("output_vis/" + str(index) + ".jpg", f)
def extract_points():
"""
Extract the points by taking -5/+5 in every direction from point center
"""
features, labels = get_dataset(None)
index = 0
for f, l in zip(features, labels):
if l.classification == 0:
continue
f = cv2.resize(
f, (f.shape[0] * RESIZE_FACTOR, f.shape[1] * RESIZE_FACTOR),
interpolation=cv2.INTER_CUBIC)
if l.classification == 1:
x_spot, y_spot = round(l.X_first_spot * RESIZE_FACTOR), round(
l.Y_first_spot * RESIZE_FACTOR)
spot1 = f[max(1, y_spot - SIZE_SPOT):min(f.shape[1], y_spot +
SIZE_SPOT),
max(0, x_spot - SIZE_SPOT):min(f.shape[0], x_spot +
SIZE_SPOT)]
cv2.imwrite("output/spots/" + str(index).zfill(7) + ".tiff", \
spot1)
index += 1
else:
x_spot1, y_spot1 = round(l.X_first_spot * RESIZE_FACTOR), round(
l.Y_first_spot * RESIZE_FACTOR)
cv2.imwrite("output/spots/" + str(index).zfill(7) + ".tiff", \
f[max(1, y_spot1 - SIZE_SPOT):min(f.shape[1], y_spot1 + SIZE_SPOT),
max(0, x_spot1 - SIZE_SPOT):min(f.shape[0], x_spot1 + SIZE_SPOT)])
x_spot2, y_spot2 = round(l.X_second_spot * RESIZE_FACTOR), round(
l.Y_second_spot * RESIZE_FACTOR)
cv2.imwrite("output/spots/" + str(index+1).zfill(7) + ".tiff", \
f[max(1, y_spot2 - SIZE_SPOT):min(f.shape[1], y_spot2 + SIZE_SPOT),
max(0, x_spot2 - SIZE_SPOT):min(f.shape[0], x_spot2 + SIZE_SPOT)])
index += 2
def classification_precision():
"""
Test the classification precision from darknet YOLO network.
"""
images_yolo = read_in()
data_training, labels = get_dataset(len(images_yolo))
metric = calculate_metric(images_yolo, labels)
return metric
def find_far_images():
images_yolo = read_in()
data_training, labels = get_dataset(len(images_yolo))
for i, (answer, truth) in enumerate(zip(images_yolo, labels)):
s = score_one_image(answer, truth)
if s > THRESHOLD_ERROR:
print("image", i, "error : ", s)
print(answer)
print(truth, "\n")
def save_hists():
features, labels = get_dataset(1000)
boudaries = 4 * RESIZE_FACTOR
for i, (f, l) in enumerate(zip(features, labels)):
plt.close()
hist, bins, _ = plt.hist(f.flatten(), bins=[i * 10 for i in range(26)])
# print(hist)
if l.classification != 0:
f = visualize_points_one_image(f, l)
plt.subplot(211), plt.imshow(f, 'gray')
plt.subplot(212), plt.plot(bins[:-1], hist)
plt.xlim([0,256])
plt.savefig("output_vis/hist" + str(i).zfill(6) + ".jpg")
def extract_false_images():
"""
Extract the points by taking -5/+5 in every direction from point center
"""
features, labels = get_dataset(None)
index = 0
for f, l in zip(features, labels):
if l.classification == 0:
for i in range(1):
f = cv2.resize(
f,
(f.shape[0] * RESIZE_FACTOR, f.shape[1] * RESIZE_FACTOR),
interpolation=cv2.INTER_CUBIC)
x = random.randint(SIZE_SPOT, f.shape[0] - SIZE_SPOT - 1)
y = random.randint(SIZE_SPOT, f.shape[0] - SIZE_SPOT - 1)
cv2.imwrite("output/false_spots/" + str(index).zfill(7) + ".tiff", \
f[x - SIZE_SPOT:x + SIZE_SPOT, y -SIZE_SPOT:y+SIZE_SPOT ])
index += 1