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 root():
filepath, label, serial = next(im_gen)
im = Image.open(filepath)
assert im.format == "GIF", im.format
assert im.n_frames == 16, im.n_frames
N = 52
w, h = im.size
images = []
M0_old = None
M_mask = np.zeros((h, w), np.uint8)
for ix in (3, 7, 11, 15):
first = (M0_old is None)
im.seek(ix)
im_row = []
M0 = np.array(im.convert("RGB"))
im_row.append(M0)
M0 = cv2.cvtColor(M0, cv2.COLOR_RGB2GRAY)
if first:
im_row.append(M0)
else:
Mdiff = cv2.subtract(M0_old, M0)
im_row.append(0xff - Mdiff)
if first:
Mt = extract_c0(M0)
else:
Mt = extract_c123(M0, M0_old)
M0_old = M0
Mt_inv = 0xff - Mt
Mt = 0xff - np.bitwise_and(Mt_inv, 0xff - M_mask)
M_mask = np.bitwise_or(M_mask, Mt_inv)
im_row.append(Mt)
Mt = crop(Mt, first)
im_row.append(Mt)
images.append(im_row)
im.close()
ctx = {
"images": images,
"label": label,
"serial": serial,
}
return render_template("label_server.html", **ctx)
def generate_dataset(dataset_dir, crop_size, img_list, label_list):
img_path = os.path.join(dataset_dir, 'img')
label_path = os.path.join(dataset_dir, 'label')
visualize_gt_path = os.path.join(dataset_dir, 'visualize_gt')
if (not os.path.exists(img_path)):
os.mkdir(img_path)
if (not os.path.exists(label_path)):
os.mkdir(label_path)
if (not os.path.exists(visualize_gt_path)):
os.mkdir(visualize_gt_path)
for i in range(len(img_list)):
crop(img_list[i],
label_list[i],
crop_size,
crop_size,
prefix='%d' % (i + 1),
save_dir=dataset_dir,
crop_label=True)
generate_trainval_list(dataset_dir)
write_train_list(dataset_dir)
def predict_captcha(im_data, model):
fp = BytesIO(im_data)
im = Image.open(fp)
assert im.format == "GIF", im.format
assert im.n_frames == 16, im.n_frames
N = CAPTCHA_SIZE
w, h = im.size
Xlist = []
ylist = []
M0_list = []
M0_last = None
M_mask = np.zeros((h, w), dtype=np.uint8)
M_merge = np.full((h, w), 0xff, dtype=np.uint8)
for ix in (3, 7, 11, 15):
im.seek(ix)
M0 = np.array(im.convert('RGB'))
M0 = cv2.cvtColor(M0, cv2.COLOR_RGB2GRAY)
M0_list.append(M0)
M_merge = np.min([M_merge, M0], axis=0)
for cix, M0 in enumerate(M0_list):
first = (M0_last is None)
if first:
Mt = extract_c0_v2(M0, M_merge)
else:
Mt = extract_c123(M0, M0_last)
M0_last = M0
Mt_inv = 0xff - Mt
Mt = 0xff - np.bitwise_and(Mt_inv, 0xff - M_mask)
M_mask = np.bitwise_or(M_mask, Mt_inv)
Mt = crop(Mt, first)
Xlist.append(Mt)
im.close()
fp.close()
Xlist = np.array(Xlist, dtype=np.float32).reshape(-1, 1, N, N)
ylist = model(torch.from_numpy(Xlist))
code = ''.join(CAPTCHA_LABELS[ix] for ix in torch.argmax(ylist, dim=1))
return code
def test():
"""
Train the model
**input: **
*dataset: (String) Dataset folder to used
*ckpt: (String) [Optional] Path to the ckpt file to restore
"""
ckpt = './outputs/checkpoints/SAR'
validation_data_dir = './data/Test'
img_width, img_height = 100, 100
# Load name of id
classname = [
'2S1', 'BMP-2', 'BRDM-2', 'BTR-70', 'BTR-60', 'D7', 'T62', 'T72',
'ZIL131', 'ZSU234'
]
# Get Test dataset
model = ModelSAR("SAR", output_folder=None)
# Load the model
model.load(ckpt)
for i in range(1):
test_datagen = ImageDataGenerator(rescale=1. / 255, rotation_range=0)
validation_generator = test_datagen.flow_from_directory(
validation_data_dir,
target_size=(img_height, img_width),
batch_size=2425,
class_mode='sparse')
filenames = validation_generator.class_indices
print(filenames)
X_test, y_test = validation_generator.next()
X_test = crop(X_test)
#X_test = blur(X_test, pow(10, 1 + i * 0.2))
#X_test = oclusion(X_test, 0.06*(1 + i ))
#X_test = noise(X_test,0.01*(i+1))
# Evaluate all the dataset
loss, acc, predicted_class = model.evaluate_dataset(
np.expand_dims(X_test[:, :, :, 1], axis=3), y_test)
print("Accuracy = ", acc)
print("Loss = ", loss)
# Get the confusion matrix
cnf_matrix = confusion_matrix(y_test, predicted_class)
# Plot the confusion matrix
plt.figure(figsize=(5.2, 5))
plot_confusion_matrix(cnf_matrix,
classes=classname,
title='Confusion matrix, without normalization')
plt.show()
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 __init__(self, T1_path, IR_path, T2_path, is_transform, is_crop,
is_hist, forest):
self.T1_path = T1_path
self.IR_path = IR_path
self.T2_path = T2_path
self.is_transform = is_transform
self.is_crop = is_crop
self.is_hist = is_hist
self.forest = forest
self.n_classes = 11
self.T1mean = 0.0
self.IRmean = 0.0
self.T2mean = 0.0
#read data
T1_nii = to_uint8(readVol(self.T1_path))
IR_nii = IR_to_uint8(readVol(self.IR_path))
T2_nii = to_uint8(readVol(self.T2_path))
#histeq
if self.is_hist:
T1_nii = histeq(T1_nii)
#stack
T1_stack_list = get_stacked(T1_nii, self.forest)
IR_stack_list = get_stacked(IR_nii, self.forest)
T2_stack_list = get_stacked(T2_nii, self.forest)
#crop
if self.is_crop:
region_list = calc_max_region_list(
calc_crop_region(T1_stack_list, 50, 5), self.forest)
self.region_list = region_list
T1_stack_list = crop(T1_stack_list, region_list)
IR_stack_list = crop(IR_stack_list, region_list)
T2_stack_list = crop(T2_stack_list, region_list)
#get mean
'''
T1mean,IRmean,T2mean=0.0,0.0,0.0
for samples in T1_stack_list:
for stacks in samples:
T1mean=T1mean+np.mean(stacks)
self.T1mean=T1mean/(len(T1_stack_list)*len(T1_stack_list[0]))
for samples in IR_stack_list:
for stacks in samples:
IRmean=IRmean+np.mean(stacks)
self.IRmean=IRmean/(len(IR_stack_list)*len(IR_stack_list[0]))
for samples in T2_stack_list:
for stacks in samples:
T2mean=T2mean+np.mean(stacks)
self.T2mean=T2mean/(len(T2_stack_list)*len(T2_stack_list[0]))
'''
self.T1mean = 94.661544495
self.IRmean = 88.574705283
self.T2mean = 32.376038631
#transform
if self.is_transform:
for stack_index in range(len(T1_stack_list)):
T1_stack_list[stack_index], \
IR_stack_list[stack_index], \
T2_stack_list[stack_index]= \
self.transform( \
T1_stack_list[stack_index], \
IR_stack_list[stack_index], \
T2_stack_list[stack_index])
# data ready
self.T1_stack_list = T1_stack_list
self.IR_stack_list = IR_stack_list
self.T2_stack_list = T2_stack_list
def train():
"""
Train the model
**input: **
*dataset: (String) Dataset folder to used
*ckpt: (String) [Optional] Path to the ckpt file to restore
*output: (String) [Optional] Path to the output folder to used. ./outputs/ by default
"""
ckpt = './outputs/checkpoints/SAR'
output_folder = None
train_data_dir = './data/Train'
validation_data_dir = './data/Train'
img_width, img_height = 100, 100
NB_LABELS = 10
nb_train_samples = 2747
batch_size = 16
lr = 0.0002
train_datagen = ImageDataGenerator(rescale=1. / 255)
valid_datagen = ImageDataGenerator(rescale=1. / 255)
train_generator = train_datagen.flow_from_directory(
train_data_dir,
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='sparse',
shuffle=True)
valid_generator = valid_datagen.flow_from_directory(
validation_data_dir,
target_size=(img_height, img_width),
batch_size=nb_train_samples,
class_mode='sparse',
shuffle=True)
# Init model
model = ModelSAR("SAR", output_folder, NB_LABELS)
if ckpt is not None:
model.init()
else:
model.load(ckpt)
# Training pipeline
for batch_id in range(40000):
x_batch, y_batch = train_generator.next()
x_batch = random_crop(x_batch)
### Training
cost, acc = model.optimize(np.expand_dims(x_batch[:, :, :, 1], axis=3),
y_batch, lr)
print("Batch ID = %s, loss = %s, acc = %s" % (batch_id, cost, acc))
### Validation
if batch_id % 20 == 0: # Plot the last results
X_valid, y_valid = valid_generator.next()
X_valid = crop(X_valid)
print("Evaluate full validation dataset ...")
loss, acc, _ = model.evaluate_dataset(
np.expand_dims(X_valid[:, :, :, 1], axis=3), y_valid)
print("Current loss: %s Current acc: %s" % (loss, acc))
model.save()