def scutfbp_df(queue):
"""Preprocess and augment SCUT-FBP faces to data/. Returns pandas dataframe"""
scutfbp_df = pd.read_excel("Rating_Collection/Attractiveness label.xlsx")
# Convert type of #Image column to str
scutfbp_df["#Image"] = scutfbp_df["#Image"].astype(str)
# Drop column Standard Deviation
scutfbp_df = scutfbp_df.drop("Standard Deviation", 1)
for face in os.listdir("Data_Collection"):
if face.endswith(".jpg"):
base = os.path.splitext(face)[0]
# Regex to find numbers at end of string
img_num = re.match(".*?([0-9]+)$", base).group(1)
try:
preprocess.resize("Data_Collection/{0}".format(face), "data/{0}".format(face))
preprocess.hflip("data/{0}".format(face), "data/{0}-F.jpg".format(base))
preprocess.add_noise("data/{0}".format(face), "data/{0}-N.jpg".format(base))
except:
scutfbp_df = scutfbp_df[getattr(scutfbp_df, "#Image") != img_num]
flipped_df = scutfbp_df.copy()
noisy_df = scutfbp_df.copy()
flipped_df["#Image"] = "SCUT-FBP-" + flipped_df["#Image"] + "-F.jpg"
noisy_df["#Image"] = "SCUT-FBP-" + noisy_df["#Image"] + "-N.jpg"
scutfbp_df["#Image"] = "SCUT-FBP-" + scutfbp_df["#Image"] + ".jpg"
df = pd.concat([scutfbp_df, flipped_df, noisy_df], ignore_index=True)
# Rename #Image -> Face and Attractiveness label -> Rating
df.columns = ["Face", "Rating"]
# Convert from 5 point scale to 10 point scale
df["Rating"] *= 10.0 / 5.0
queue.put(df)
def models_df(queue):
"""Preprocess and augment models.com faces to data/. Returns pandas dataframe"""
imgs = []
for i in range(1, 216):
url = "http://models.com/newfaces/page/{0}".format(i)
page = urllib2.urlopen(url).read()
soup = BeautifulSoup(page, "lxml")
for tag in soup.findAll("img", {"class": "attachment-square"}):
src = "http:{0}".format(tag["src"])
base = uuid.uuid4().hex
filename = base + ".jpg"
urllib.urlretrieve(src, "data/{0}".format(filename))
try:
preprocess.resize("data/{0}".format(filename), "data/{0}".format(filename))
preprocess.hflip("data/{0}".format(filename), "data/{0}-F.jpg".format(base))
preprocess.add_noise("data/{0}".format(filename), "data/{0}-N.jpg".format(base))
except:
os.remove("data/{0}".format(filename))
continue
imgs.append({"Face": filename, "Rating": 10})
imgs.append({"Face": "{0}-F.jpg".format(base), "Rating": 10})
imgs.append({"Face": "{0}-N.jpg".format(base), "Rating": 10})
df = pd.DataFrame(imgs)
queue.put(df)
def usfaces_df(queue):
"""Preprocess and augment US Face Database faces to data/. Returns pandas dataframe"""
usfaces_df = pd.read_excel("Full Attribute Scores/demographic & others labels/demographic-others-labels.xlsx")
usfaces_df = usfaces_df[["Filename", "Attractive"]]
usfaces_df = usfaces_df.drop_duplicates(["Filename"])
for face in usfaces_df["Filename"]:
base = os.path.splitext(face)[0]
try:
preprocess.resize("10k US Adult Faces Database/Face Images/{0}".format(face), "data/{0}".format(face))
preprocess.hflip("data/{0}".format(face), "data/{0}-F.jpg".format(base))
preprocess.add_noise("data/{0}".format(face), "data/{0}-N.jpg".format(base))
except:
usfaces_df = usfaces_df[usfaces_df.Filename != face]
flipped_df = usfaces_df.copy()
noisy_df = usfaces_df.copy()
flipped_df["Filename"] = flipped_df["Filename"].str[:-4] + "-F.jpg"
noisy_df["Filename"] = noisy_df["Filename"].str[:-4] + "-N.jpg"
df = pd.concat([usfaces_df, flipped_df, noisy_df], ignore_index=True)
df.columns = ["Face", "Rating"]
df["Rating"] *= 10.0 / 5.0
queue.put(df)
def main() -> None:
(x_train_s, x_ref, y_ref), _ = dataset.get_fasion_mnist()
x_train_s = preprocess.resize(x_train_s)
x_ref = preprocess.resize(x_ref)
np.random.seed(0)
train(x_train_s, x_ref, y_ref, 40)
def eccv_df(queue):
"""Preprocess and augment Gray et al. dataset to data/. Returns pandas dataframe"""
root = ET.parse("eccv2010_beauty_data/hotornot_face_all.xml").getroot()
childs = []
for child in root:
filename = os.path.split(child.attrib["filename"])[-1]
base = os.path.splitext(filename)[0]
try:
preprocess.resize(
"eccv2010_beauty_data/{0}".format(child.attrib["filename"]),
"data/{0}".format(base + ".jpg"),
crop=False,
)
preprocess.hflip("data/{0}".format(filename), "data/{0}-F.jpg".format(base))
preprocess.add_noise("data/{0}".format(filename), "data/{0}-N.jpg".format(base))
except:
continue
childs.append([base + "-F.jpg", float(child.attrib["score"])])
childs.append([base + "-N.jpg", float(child.attrib["score"])])
childs.append([base + ".jpg", float(child.attrib["score"])])
df = pd.DataFrame(childs, columns=["Face", "Rating"])
df["Rating"] += 4
df["Rating"] *= 10.0 / 8.0
queue.put(df)
def chicago_df(queue):
"""Preprocess and augment Chicago faces to data/. Returns pandas dataframe"""
chicago_df = pd.read_excel("CFD Version 2.0/CFD 2.0 Norming Data and Codebook.xlsx", skiprows=4)
chicago_df = chicago_df[["Target", "Attractive"]]
for dir in os.listdir("CFD Version 2.0/CFD 2.0 Images"):
if dir == ".DS_Store":
continue
for face in os.listdir("CFD Version 2.0/CFD 2.0 Images/{0}".format(dir)):
# Neutral faces
if face.endswith("N.jpg"):
# Is one face detected
try:
preprocess.resize(
"CFD Version 2.0/CFD 2.0 Images/{0}/{1}".format(dir, face), "data/{0}.jpg".format(dir)
)
preprocess.hflip("data/{0}.jpg".format(dir), "data/{0}-F.jpg".format(dir))
preprocess.add_noise("data/{0}.jpg".format(dir), "data/{0}-N.jpg".format(dir))
except:
chicago_df = chicago_df[chicago_df.Target != dir]
flipped_df = chicago_df.copy()
noisy_df = chicago_df.copy()
flipped_df["Target"] = flipped_df["Target"] + "-F.jpg"
noisy_df["Target"] = noisy_df["Target"] + "-N.jpg"
chicago_df["Target"] = chicago_df["Target"] + ".jpg"
df = pd.concat([chicago_df, flipped_df, noisy_df], ignore_index=True)
# Rename Target -> Face and Attractive -> Rating
df.columns = ["Face", "Rating"]
# Convert from 7 point scale to 10 point scale
df["Rating"] *= 10.0 / 7.0
queue.put(df)
def main() -> None:
(x_train_s, _, _), (x_test_s, x_test_b) = dataset.get_fasion_mnist()
x_train_s = preprocess.resize(x_train_s)
x_test_s = preprocess.resize(x_test_s)
x_test_b = preprocess.resize(x_test_b)
network = tf.keras.models.load_model(
"_data/model.h5", custom_objects={"original_loss": original_loss})
network.summary()
predict(x_train_s, x_test_s, x_test_b, network)
def load_image_test(image_file):
input_image, real_image = load(image_file)
input_image, real_image = resize(input_image, real_image, IMG_HEIGHT,
IMG_WIDTH)
input_image, real_image = normalize(input_image, real_image)
return input_image, real_image
def telemetry(sid, data):
if data:
# The current steering angle of the car
steering_angle = data["steering_angle"]
# The current throttle of the car
throttle = data["throttle"]
# The current speed of the car
speed = data["speed"]
# The current image from the center camera of the car
imgString = data["image"]
image = Image.open(BytesIO(base64.b64decode(imgString)))
image_array = np.asarray(image)
image_array = preprocess.crop(image_array, 0.35, 0.1)
image_array = preprocess.resize(image_array, new_dimension=(64, 64))
transformed_image_array = image_array[None, :, :, :]
steering_angle = float(
model.predict(transformed_image_array, batch_size=1))
throttle = controller.update(float(speed))
print(steering_angle, throttle)
send_control(steering_angle, throttle)
# save frame
if args.image_folder != '':
timestamp = datetime.utcnow().strftime('%Y_%m_%d_%H_%M_%S_%f')[:-3]
image_filename = os.path.join(args.image_folder, timestamp)
image.save('{}.jpg'.format(image_filename))
else:
# NOTE: DON'T EDIT THIS.
sio.emit('manual', data={}, skip_sid=True)
def do_preprocess(self, img):
x = np.copy(img)
if self.properties["autocrop"]["enabled"]:
x = preprocess.autocrop(x)
if x is None:
return None
x = preprocess.blur(
x, gaussian_blur=self.properties["gaussian_blur"], median_blur=self.properties["median_blur"]
)
if self.properties["grey"]["enabled"]:
x = preprocess.grey(x)
if self.properties["resize"]["enabled"]:
x = preprocess.resize(x, (self.properties["resize"]["width"], self.properties["resize"]["height"]))
elif self.properties["scale_max"]["enabled"]:
x = preprocess.scale_max(x, self.properties["scale_max"]["width"], self.properties["scale_max"]["height"])
if self.properties["convert_to_matrix_colors"]["enabled"]:
(matrix, cluster_centers_, labels, background_label) = improc.color.Matrix_scikit_kmeans(
x, self.properties["convert_to_matrix_colors"]["number_of_colors"]
)
x = improc.color.Image_from_matrix(
matrix,
self.properties["convert_to_matrix_colors"]["height"],
self.properties["convert_to_matrix_colors"]["width"],
)
return x
def do_preprocess(self, img):
x = np.copy(img)
if x is not None and self.properties["autocrop"]["enabled"]:
x = preprocess.autocrop(x)
if x is not None:
x = preprocess.blur(
x, gaussian_blur=self.properties["gaussian_blur"], median_blur=self.properties["median_blur"]
)
if x is not None and self.properties["grey"]["enabled"]:
x = preprocess.grey(x)
if x is not None and self.properties["bitwise"]["enabled"]:
x = preprocess.bitwise(x)
if x is not None and self.properties["canny"]["enabled"]:
x = preprocess.canny(x, self.properties["canny"]["threshold1"], self.properties["canny"]["threshold2"])
if x is not None and self.properties["laplacian"]["enabled"]:
x = preprocess.laplacian(x)
if x is not None and self.properties["thresh"]["enabled"]:
x = preprocess.thresh(x)
if x is not None and self.properties["closing"]["enabled"]:
x = preprocess.closing(x, self.properties["closing"]["width"], self.properties["closing"]["height"])
if x is not None and self.properties["dilate"]["enabled"]:
x = preprocess.dilate(
x,
self.properties["dilate"]["width"],
self.properties["dilate"]["height"],
self.properties["dilate"]["iterations"],
)
if x is not None and self.properties["outline_contour"]["enabled"]:
x = preprocess.outline_contour(x)
if x is not None and self.properties["resize"]["enabled"]:
x = preprocess.resize(x, (self.properties["resize"]["width"], self.properties["resize"]["height"]))
elif x is not None and self.properties["scale_max"]["enabled"]:
x = preprocess.scale_max(x, self.properties["scale_max"]["width"], self.properties["scale_max"]["height"])
if x is not None and self.properties["add_border"]["enabled"]:
x = preprocess.add_border(
x,
border_size=self.properties["add_border"]["border_size"],
color_value=self.properties["add_border"]["color_value"],
fill_dimensions=self.properties["add_border"]["fill_dimensions"],
)
return x
def do_preprocess(self, img):
x = np.copy(img)
if self.properties["autocrop"]["enabled"]:
x = preprocess.autocrop(x)
if x is None:
return None
x = preprocess.blur(
x,
gaussian_blur=self.properties["gaussian_blur"],
median_blur=self.properties["median_blur"]
)
if self.properties["grey"]["enabled"]:
x = preprocess.grey(x)
if self.properties["resize"]["enabled"]:
x = preprocess.resize(
x,
(
self.properties["resize"]["width"],
self.properties["resize"]["height"]
)
)
elif self.properties["scale_max"]["enabled"]:
x = preprocess.scale_max(
x,
self.properties["scale_max"]["width"],
self.properties["scale_max"]["height"]
)
if self.properties["convert_to_matrix_colors"]["enabled"]:
(
matrix, cluster_centers_, labels, background_label
) = improc.color.Matrix_scikit_kmeans(
x,
self.properties["convert_to_matrix_colors"]["number_of_colors"]
)
x = improc.color.Image_from_matrix(
matrix,
self.properties["convert_to_matrix_colors"]["height"],
self.properties["convert_to_matrix_colors"]["width"]
)
return x
def preprocess(im, augment=False):
with tf.variable_scope("preprocess"):
if augment:
im = pp.random_resize(im, INPUT_SIZE)
im = pp.random_flip(im)
im = pp.random_distort_color(im)
if GRAYSCALE: im = pp.grayscale(im)
im = pp.random_op(im, [
lambda _: _, pp.sharpen, pp.avg_blur, pp.unsharpen,
pp.gauss_blur
])
else:
im = pp.resize(im, INPUT_SIZE)
if GRAYSCALE: im = pp.grayscale(im)
im = pp.normalize(im, 0.5, 0.5)
return im
def telemetry(sid, data):
# The current steering angle of the car
steering_angle = data["steering_angle"]
# The current throttle of the car
throttle = data["throttle"]
# The current speed of the car
speed = data["speed"]
# The current image from the center camera of the car
imgString = data["image"]
image = Image.open(BytesIO(base64.b64decode(imgString)))
image_array = np.asarray(image)
image_array = preprocess.crop(image_array, 0.35, 0.1)
image_array = preprocess.resize(image_array, new_dim=(64, 64))
transformed_image_array = image_array[None, :, :, :]
# This model currently assumes that the features of the model are just the images. Feel free to change this.
steering_angle = float(model.predict(transformed_image_array,
batch_size=1))
# The driving model currently just outputs a constant throttle. Feel free to edit this.
#throttle = 0.3
throttle = 0.2
if float(speed) < 10 and float(speed) > 5:
throttle = 0.5
print('{:.5f}, {:.1f}'.format(steering_angle, throttle))
send_control(steering_angle, throttle)
# save frame
if args.image_folder != '':
timestamp = datetime.utcnow().strftime('%Y_%m_%d_%H_%M_%S_%f')[:-3]
image_filename = os.path.join(args.image_folder, timestamp)
image.save('{}.jpg'.format(image_filename))
def runBatch(video_reader, gps_dat, cam, output_base, start_frame, final_frame, left_lanes, right_lanes, ldist, rdist, tr):
meters_per_point = 10
points_fwd = 8
frames_per_second = 50
distances = GPSVelocities(gps_dat) / frames_per_second
pts = GPSPos(gps_dat, cam, gps_dat[0, :]) # points wrt camera
count = 0
output_num = 0
imgs = []
labels = []
success = True
while True:
(success, I, frame, P) = video_reader.getNextFrame()
#P = np.eye(3)
if count % 160 == 0:
print count
if success == False or frame >= tr.shape[0]:
print success, frame, tr.shape[0]
break
if frame < start_frame or (final_frame != -1 and frame >= final_frame):
continue
important_frames = (outputDistances(distances, frame, meters_per_point, points_fwd, 2))
if len(important_frames) < points_fwd:
continue
important_left = []
important_right = []
velocities = gps_dat[important_frames,4:7]
velocities[:,[0, 1]] = velocities[:,[1, 0]]
sideways = np.cross(velocities, np.array([0,0,1]), axisa=1)
sideways/= np.sqrt((sideways ** 2).sum(-1))[..., np.newaxis]
vel_start = ENU2IMU(np.transpose(velocities), gps_dat[0,:])
vel_current = ENU2IMU(np.transpose(velocities), gps_dat[frame,:])
sideways_start = GPSPosCamera(np.cross(vel_start, np.array([0,0,1]), axisa=0).transpose(), cam) # sideways vector wrt starting frame (camera)
sideways_current = GPSPosCamera(np.cross(vel_current, np.array([0,0,1]), axisa=0).transpose(), cam) # sideways vector wrt starting frame (camera)
sideways_start /= np.sqrt((sideways_start ** 2).sum(0))[np.newaxis, ...]
sideways_current /= np.sqrt((sideways_current ** 2).sum(0))[np.newaxis, ...] # save sideways_current
center = GPSPos(gps_dat[important_frames,:], cam, gps_dat[0, :]) # points wrt imu
for ind in xrange(len(important_frames)):
fr = important_frames[ind]
min_val = max(fr - 50, 0)
max_val = min(fr + 50, left_lanes.shape[0] - 1)
l_distances = np.cross(left_lanes[min_val:max_val, 0:3] - np.transpose(center[:,ind]), np.transpose(sideways_start[:, ind]), axisa=1)
l_distances = np.sqrt((l_distances ** 2).sum(-1))
r_distances = np.cross(right_lanes[min_val:max_val, 0:3] - np.transpose(center[:,ind]), np.transpose(sideways_start[:, ind]), axisa=1)
r_distances = np.sqrt((r_distances ** 2).sum(-1))
important_left.append(np.argmin(l_distances)+min_val)
important_right.append(np.argmin(r_distances)+min_val)
important_left = np.array(important_left)
important_right = np.array(important_right)
max_idx = max(max(np.max(important_frames), np.max(important_left)), np.max(important_right))
if max_idx >= left_lanes.shape[0] or max_idx >= right_lanes.shape[0]:
print 'maxing out'
continue
temp_left = np.linalg.solve(tr[frame, :, :], left_lanes[important_left, :].transpose()) # save temp_left [0:3,:]
#temp_left[0:3, :] += sideways_current*0.8
temp_gps = GPSPos(gps_dat[important_frames], cam, gps_dat[frame, :]) # save temp_gps [0:3,:]
temp_right = np.linalg.solve(tr[frame, :, :], right_lanes[important_right, :].transpose()) # save temp_right [0:3,:]
#temp_right[0:3, :] -= sideways_current*0.8
outputs = []
for i in xrange(temp_left.shape[1]):
outputs.append(temp_left[0, i])
outputs.append(temp_left[1, i])
outputs.append(temp_left[2, i])
for i in xrange(temp_gps.shape[1]):
outputs.append(temp_gps[0, i])
outputs.append(temp_gps[1, i])
outputs.append(temp_gps[2, i])
for i in xrange(temp_left.shape[1]):
outputs.append(temp_right[0, i])
outputs.append(temp_right[1, i])
outputs.append(temp_right[2, i])
for i in xrange(sideways_current.shape[1]):
outputs.append(sideways_current[0, i])
outputs.append(sideways_current[1, i])
outputs.append(sideways_current[2, i])
for i in xrange(P.shape[0]):
for j in xrange(P.shape[1]):
outputs.append(P[i, j])
labels.append(outputs)
reshaped = pp.resize(I, (240, 320))[0]
imgs.append(reshaped)
if len(imgs) == 960:
merge_file = "%s_%d" % (output_base, output_num)
pml.save_merged_file(merge_file, imgs, labels, imgRows=len(labels[0]))
imgs = []
labels = []
output_num += 1
count += 1
dataset_name = "stanford_dogs"
(ds_train, ds_val, ds_test), ds_info = tfds.load(dataset_name,
split=["train", "test[:50%]", "test[50%:100%]"],
with_info=True,
as_supervised=True)
# 犬の種類数
NUM_CLASSES = ds_info.features["label"].num_classes
names = ds_info.features["label"].names
IMG_SIZE = 300
BATCH_SIZE = 100
size = (IMG_SIZE, IMG_SIZE)
ds_train, ds_val, ds_test = preprocess.resize(train=ds_train,
val=ds_val,
test=ds_test,
size=size)
ds_train, ds_val, ds_test = preprocess.batch_create(train=ds_train,
val=ds_val,
test=ds_test,
NUM_CLASSES=NUM_CLASSES,
BATCH_SIZE=BATCH_SIZE,
BUFFER_SIZE=len(ds_test))
model, lr = post_process.build_model(NUM_CLASSES=NUM_CLASSES, IMG_SIZE=IMG_SIZE)
model.load_weights(post_process.checkpoint(num=0, IMG_SIZE=IMG_SIZE))
# es_callback = post_process.early_stopping()
def runBatch(video_reader, gps_dat, cam, output_base, start_frame, final_frame, left_lanes, right_lanes, ldist, rdist, tr):
meters_per_point = 10
points_fwd = 8
frames_per_second = 50
distances = GPSVelocities(gps_dat) / frames_per_second
pts = GPSPos(gps_dat, cam, gps_dat[0, :]) # points wrt camera
count = 0
output_num = 0
imgs = []
labels = []
success = True
while True:
#(success, I, frame, P) = video_reader.getNextFrame()
(success, I) = video_reader.getNextFrame()
frame = 10*count
#frame = count
P = np.eye(3)
if count % 160 == 0:
print count
if success == False or frame >= tr.shape[0]:
print success, frame, tr.shape[0]
break
if frame < start_frame or (final_frame != -1 and frame >= final_frame):
continue
important_frames = (outputDistances(distances, frame, meters_per_point, points_fwd, 2))
if len(important_frames) < points_fwd:
continue
important_left = []
important_right = []
velocities = gps_dat[important_frames,4:7]
velocities[:,[0, 1]] = velocities[:,[1, 0]]
sideways = np.cross(velocities, np.array([0,0,1]), axisa=1)
sideways/= np.sqrt((sideways ** 2).sum(-1))[..., np.newaxis]
vel_start = ENU2IMU(np.transpose(velocities), gps_dat[0,:])
vel_current = ENU2IMU(np.transpose(velocities), gps_dat[frame,:])
sideways_start = GPSPosCamera(np.cross(vel_start, np.array([0,0,1]), axisa=0).transpose(), cam) # sideways vector wrt starting frame (camera)
sideways_current = GPSPosCamera(np.cross(vel_current, np.array([0,0,1]), axisa=0).transpose(), cam) # sideways vector wrt starting frame (camera)
sideways_start /= np.sqrt((sideways_start ** 2).sum(0))[np.newaxis, ...]
sideways_current /= np.sqrt((sideways_current ** 2).sum(0))[np.newaxis, ...] # save sideways_current
center = GPSPos(gps_dat[important_frames,:], cam, gps_dat[0, :]) # points wrt imu
for ind in xrange(len(important_frames)):
fr = important_frames[ind]
min_val = max(fr - 50, 0)
max_val = min(fr + 50, left_lanes.shape[0] - 1)
l_distances = np.cross(left_lanes[min_val:max_val, 0:3] - np.transpose(center[:,ind]), np.transpose(sideways_start[:, ind]), axisa=1)
l_distances = np.sqrt((l_distances ** 2).sum(-1))
r_distances = np.cross(right_lanes[min_val:max_val, 0:3] - np.transpose(center[:,ind]), np.transpose(sideways_start[:, ind]), axisa=1)
r_distances = np.sqrt((r_distances ** 2).sum(-1))
important_left.append(np.argmin(l_distances)+min_val)
important_right.append(np.argmin(r_distances)+min_val)
important_left = np.array(important_left)
important_right = np.array(important_right)
max_idx = max(max(np.max(important_frames), np.max(important_left)), np.max(important_right))
if max_idx >= left_lanes.shape[0] or max_idx >= right_lanes.shape[0]:
print 'maxing out'
continue
temp_left = np.linalg.solve(tr[frame, :, :], left_lanes[important_left, :].transpose()) # save temp_left [0:3,:]
#temp_left[0:3, :] += sideways_current*0.8
temp_gps = GPSPos(gps_dat[important_frames], cam, gps_dat[frame, :]) # save temp_gps [0:3,:]
temp_right = np.linalg.solve(tr[frame, :, :], right_lanes[important_right, :].transpose()) # save temp_right [0:3,:]
#temp_right[0:3, :] -= sideways_current*0.8
#gps_vals = warpPoints(P, PointsMask(temp_gps[0:3, :], cam)[0:2]) # save P
#left_vals = warpPoints(P, PointsMask(temp_left[0:3, :], cam)[0:2])
#right_vals = warpPoints(P, PointsMask(temp_right[0:3, :], cam)[0:2])
#gps_vals = (gps_vals / 4).astype(np.int32)
#left_vals = (left_vals / 4).astype(np.int32)
#right_vals = (right_vals / 4).astype(np.int32)
#left_vals = left_vals.clip(0,500)
#right_vals = right_vals.clip(0,500)
#gps_vals = gps_vals.clip(0,500)
#gps_vals[0, gps_vals[0, :] >= 320] = 319
#gps_vals[1, gps_vals[1, :] >= 240] = 239
#left_vals[0, left_vals[0, :] >= 320] = 319
#left_vals[1, left_vals[1, :] >= 240] = 239
#right_vals[0, right_vals[0, :] >= 320] = 319
#right_vals[1, right_vals[1, :] >= 240] = 239
#outputs = []
# scale down column numbers by 16 to aid bucketing but only scale down
# row numbers by 4 to aid visualization
#for i in xrange(points_fwd):
# outputs.append(left_vals[0, i] / 4)
# outputs.append(gps_vals[0, i] / 4)
# outputs.append(right_vals[0, i] / 4)
# outputs.append(left_vals[1, i])
# outputs.append(gps_vals[1, i])
# outputs.append(right_vals[1, i])
#labels.append(outputs)
outputs = []
for i in xrange(temp_left.shape[1]):
outputs.append(temp_left[0, i])
outputs.append(temp_left[1, i])
outputs.append(temp_left[2, i])
for i in xrange(temp_gps.shape[1]):
outputs.append(temp_gps[0, i])
outputs.append(temp_gps[1, i])
outputs.append(temp_gps[2, i])
for i in xrange(temp_left.shape[1]):
outputs.append(temp_right[0, i])
outputs.append(temp_right[1, i])
outputs.append(temp_right[2, i])
for i in xrange(sideways_current.shape[1]):
outputs.append(sideways_current[0, i])
outputs.append(sideways_current[1, i])
outputs.append(sideways_current[2, i])
for i in xrange(P.shape[0]):
for j in xrange(P.shape[1]):
outputs.append(P[i, j])
labels.append(outputs)
reshaped = pp.resize(I, (240, 320))[0]
imgs.append(reshaped)
if len(imgs) == 960:
merge_file = "%s_%d" % (output_base, output_num)
pml.save_merged_file(merge_file, imgs, labels, imgRows=len(labels[0]))
imgs = []
labels = []
output_num += 1
count += 1
print("Percent of Live Birth Classified as Non-Pregnancies: {}% ".format(
100 * sum(np.logical_and(y == 2, yhat == 0)) / sum(y == 2)))
print("Percent of Live Birth Classified as SAB: {}%".format(
100 * sum(np.logical_and(y == 2, yhat == 1)) / sum(y == 2)))
print("\n")
print("Pregnancy Accuracy (SAB or Live Birth): {}%".format(
100 * sum(np.logical_and(y != 0, yhat != 0)) / sum(y != 0)))
##MODEL RUNNING
#Preprocesses the image data
all_images, names = preprocess.readImages(
'/Users/jaredgeller/Desktop/Work/Stanford/Year 3/Quarter 2/IVF_Project/part_1/'
)
all_images = preprocess.resize(all_images, 224)
namesDf = preprocess.processName(names)
sheetDf = preprocess.processXLSXData(
"/Users/jaredgeller/Desktop/Work/Stanford/Year 3/Quarter 2/IVF_Project/2018 FRESH TRANSFERS NAMES REMOVED.xlsx"
)
fullDf, y = preprocess.mergeData(namesDf, sheetDf)
#Gets training set (first 200 images)
X_train = all_images[:200]
Y_train = y[:200]
X_test = all_images[200:]
Y_test = y[200:]
#Train(X_train=X_train, Y_train=Y_train, alpha=0.001, n_epochs=100, print_every=1, batch_size=32, is_training=True)
Test(X_test=all_images, Y_test=y)
for r in each:
for c in r:
bordered_each[rowInd][colInd] = c
colInd += 1
colInd = borderSize
rowInd += 1
return bordered_each
print("printing chars: \n")
numberPlate = []
for each in segmentation.characters:
each = np.floor(255 * each)
row, col = each.shape[:2]
each = borderize(each, 5)
each = preprocess.resize(each, (40, 40), anti_aliasing_sigma=True)
each = np.floor(each)
each = abs(each - 255)
plot_image(each)
each = np.reshape(each, 1600)
numberPlate.append(one_time_test(each))
rightplate_string = ''
column_list_copy = segmentation.column_list[:]
segmentation.column_list.sort()
for each in segmentation.column_list:
rightplate_string += numberPlate[column_list_copy.index(each)]
segmentation.characters = []
segmentation.column_list = []
print(rightplate_string)
if __name__ == "__main__":
dictionary = {}
start_time = time.time()
spell.create_dictionary(dictionary,"./all_medical_terms.txt")
run_time = time.time() - start_time
print '%.2f seconds to run' % run_time
filename = parser.parse_args().filename
#Load image and change channels
image = cv2.cvtColor(cv2.imread(filename), cv2.COLOR_BGR2RGB)
#Pre-processing = Cropping + Binarization
imageEdges = pp.edgesDet(image, 200, 250)
closedEdges = cv2.morphologyEx(imageEdges, cv2.MORPH_CLOSE, np.ones((5, 11)))
pageContour = pp.findPageContours(closedEdges, pp.resize(image))
pageContour = pageContour.dot(pp.ratio(image))
newImage = pp.perspImageTransform(image, pageContour)
#Saving image to show status in the app
save_filename = filename[:-4]+"_1"+filename[-4:]
cv2.imwrite(save_filename, cv2.cvtColor(newImage, cv2.COLOR_BGR2RGB))
##Detect words using google-ocr
if (parser.parse_args().google_ocr):
entities,bBoxes = go.convert(save_filename)
detected_filename = "input.txt"
with open(detected_filename, 'w') as outfile:
for i in entities:
outfile.write(i)
outfile.write("\n")
def do_preprocess(self, img):
x = np.copy(img)
if x is not None and self.properties["autocrop"]["enabled"]:
x = preprocess.autocrop(x)
if x is not None:
x = preprocess.blur(
x,
gaussian_blur=self.properties["gaussian_blur"],
median_blur=self.properties["median_blur"]
)
if x is not None and self.properties["grey"]["enabled"]:
x = preprocess.grey(x)
if x is not None and self.properties["bitwise"]["enabled"]:
x = preprocess.bitwise(x)
if x is not None and self.properties["canny"]["enabled"]:
x = preprocess.canny(
x,
self.properties["canny"]["threshold1"],
self.properties["canny"]["threshold2"]
)
if x is not None and self.properties["laplacian"]["enabled"]:
x = preprocess.laplacian(x)
if x is not None and self.properties["thresh"]["enabled"]:
x = preprocess.thresh(x)
if x is not None and self.properties["closing"]["enabled"]:
x = preprocess.closing(
x,
self.properties["closing"]["width"],
self.properties["closing"]["height"]
)
if x is not None and self.properties["dilate"]["enabled"]:
x = preprocess.dilate(
x,
self.properties["dilate"]["width"],
self.properties["dilate"]["height"],
self.properties["dilate"]["iterations"]
)
if x is not None and self.properties["outline_contour"]["enabled"]:
x = preprocess.outline_contour(x)
if x is not None and self.properties["resize"]["enabled"]:
x = preprocess.resize(
x,
(
self.properties["resize"]["width"],
self.properties["resize"]["height"]
)
)
elif x is not None and self.properties["scale_max"]["enabled"]:
x = preprocess.scale_max(
x,
self.properties["scale_max"]["width"],
self.properties["scale_max"]["height"]
)
if x is not None and self.properties["add_border"]["enabled"]:
x = preprocess.add_border(
x,
border_size=self.properties["add_border"]["border_size"],
color_value=self.properties["add_border"]["color_value"],
fill_dimensions=self.properties["add_border"]["fill_dimensions"]
)
return x