def get_model_by_type(model_type, cfg):
from donkeycar.parts.keras import KerasRNN_LSTM, KerasBehavioral, KerasCategorical, KerasIMU, KerasLinear, Keras3D_CNN, KerasLocalizer, KerasLatent
if model_type is None:
model_type = "categorical"
input_shape = (cfg.IMAGE_H, cfg.IMAGE_W, cfg.IMAGE_DEPTH)
roi_crop = (cfg.ROI_CROP_TOP, cfg.ROI_CROP_BOTTOM)
if model_type == "localizer" or cfg.TRAIN_LOCALIZER:
kl = KerasLocalizer(num_outputs=2, num_behavior_inputs=len(cfg.BEHAVIOR_LIST), num_locations=cfg.NUM_LOCATIONS, input_shape=input_shape)
elif model_type == "behavior" or cfg.TRAIN_BEHAVIORS:
kl = KerasBehavioral(num_outputs=2, num_behavior_inputs=len(cfg.BEHAVIOR_LIST), input_shape=input_shape)
elif model_type == "imu":
kl = KerasIMU(num_outputs=2, num_imu_inputs=6, input_shape=input_shape)
elif model_type == "linear":
kl = KerasLinear(input_shape=input_shape, roi_crop=roi_crop)
elif model_type == "3d":
kl = Keras3D_CNN(image_w=cfg.IMAGE_W, image_h=cfg.IMAGE_H, image_d=cfg.IMAGE_DEPTH, seq_length=cfg.SEQUENCE_LENGTH)
elif model_type == "rnn":
kl = KerasRNN_LSTM(image_w=cfg.IMAGE_W, image_h=cfg.IMAGE_H, image_d=cfg.IMAGE_DEPTH, seq_length=cfg.SEQUENCE_LENGTH)
elif model_type == "categorical":
kl = KerasCategorical(input_shape=input_shape, throttle_range=cfg.MODEL_CATEGORICAL_MAX_THROTTLE_RANGE, roi_crop=roi_crop)
elif model_type == "latent":
kl = KerasLatent(input_shape=input_shape)
else:
raise Exception("unknown model type: %s" % model_type)
return kl
def get_model_by_type(model_type, cfg):
from donkeycar.parts.keras import KerasRNN_LSTM, KerasBehavioral, KerasCategorical, KerasIMU, KerasLinear, Keras3D_CNN
if model_type is None:
model_type = "categorical"
input_shape = (cfg.IMAGE_H, cfg.IMAGE_W, cfg.IMAGE_DEPTH)
if model_type == "behavior" or cfg.TRAIN_BEHAVIORS:
kl = KerasBehavioral(num_outputs=2,
num_behavior_inputs=len(cfg.BEHAVIOR_LIST),
input_shape=input_shape)
elif model_type == "imu":
kl = KerasIMU(num_outputs=2, num_imu_inputs=6, input_shape=input_shape)
elif model_type == "linear":
kl = KerasLinear(input_shape=input_shape)
elif model_type == "3d":
kl = Keras3D_CNN(image_w=cfg.IMAGE_W,
image_h=cfg.IMAGE_H,
image_d=cfg.IMAGE_DEPTH,
seq_length=cfg.SEQUENCE_LENGTH)
elif model_type == "rnn":
kl = KerasRNN_LSTM(seq_length=cfg.SEQUENCE_LENGTH,
input_shape=input_shape)
elif model_type == "categorical":
kl = KerasCategorical(input_shape=input_shape)
else:
raise Exception("unknown model type: %s" % model_type)
return kl
def get_model_by_type(model_type, cfg):
'''
given the string model_type and the configuration settings in cfg
create a Keras model and return it.
'''
from donkeycar.parts.keras import KerasRNN_LSTM, KerasBehavioral, KerasCategorical, KerasIMU, KerasLinear, Keras3D_CNN, KerasLocalizer, KerasLatent
from donkeycar.parts.tflite import TFLitePilot
if model_type is None:
model_type = cfg.DEFAULT_MODEL_TYPE
print("\"get_model_by_type\" model Type is: {}".format(model_type))
input_shape = (cfg.IMAGE_H, cfg.IMAGE_W, cfg.IMAGE_DEPTH)
roi_crop = (cfg.ROI_CROP_TOP, cfg.ROI_CROP_BOTTOM)
if model_type == "tflite_linear":
kl = TFLitePilot()
elif model_type == "localizer" or cfg.TRAIN_LOCALIZER:
kl = KerasLocalizer(num_outputs=2,
num_behavior_inputs=len(cfg.BEHAVIOR_LIST),
num_locations=cfg.NUM_LOCATIONS,
input_shape=input_shape)
elif model_type == "behavior" or cfg.TRAIN_BEHAVIORS:
kl = KerasBehavioral(num_outputs=2,
num_behavior_inputs=len(cfg.BEHAVIOR_LIST),
input_shape=input_shape)
elif model_type == "imu":
kl = KerasIMU(num_outputs=2, num_imu_inputs=6, input_shape=input_shape)
elif model_type == "linear":
kl = KerasLinear(input_shape=input_shape, roi_crop=roi_crop)
elif model_type == "tensorrt_linear":
# Aggressively lazy load this. This module imports pycuda.autoinit which causes a lot of unexpected things
# to happen when using TF-GPU for training.
from donkeycar.parts.tensorrt import TensorRTLinear
kl = TensorRTLinear(cfg=cfg)
elif model_type == "3d":
kl = Keras3D_CNN(image_w=cfg.IMAGE_W,
image_h=cfg.IMAGE_H,
image_d=cfg.IMAGE_DEPTH,
seq_length=cfg.SEQUENCE_LENGTH)
elif model_type == "rnn":
kl = KerasRNN_LSTM(image_w=cfg.IMAGE_W,
image_h=cfg.IMAGE_H,
image_d=cfg.IMAGE_DEPTH,
seq_length=cfg.SEQUENCE_LENGTH)
elif model_type == "categorical":
kl = KerasCategorical(
input_shape=input_shape,
throttle_range=cfg.MODEL_CATEGORICAL_MAX_THROTTLE_RANGE,
roi_crop=roi_crop)
elif model_type == "latent":
kl = KerasLatent(input_shape=input_shape)
else:
raise Exception("unknown model type: %s" % model_type)
return kl
def train(cfg, tub_names, model_name):
'''
use the specified data in tub_names to train an artifical neural network
saves the output trained model as model_name
'''
X_keys = ['cam/image_array', 'imu_array']
y_keys = ['user/angle', 'user/throttle']
def rt(rec):
rec['imu_array'] = np.array([
rec['imu/acl_x'], rec['imu/acl_y'], rec['imu/acl_z'],
rec['imu/gyr_x'], rec['imu/gyr_y'], rec['imu/gyr_z'],
rec['imu/temp']
])
return rec
kl = KerasIMU()
print('tub_names', tub_names)
if not tub_names:
tub_names = os.path.join(cfg.DATA_PATH, '*')
tubgroup = TubGroup(tub_names)
train_gen, val_gen = tubgroup.get_train_val_gen(
X_keys,
y_keys,
record_transform=rt,
batch_size=cfg.BATCH_SIZE,
train_frac=cfg.TRAIN_TEST_SPLIT)
model_path = os.path.expanduser(model_name)
total_records = len(tubgroup.df)
total_train = int(total_records * cfg.TRAIN_TEST_SPLIT)
total_val = total_records - total_train
print('train: %d, validation: %d' % (total_train, total_val))
steps_per_epoch = total_train // cfg.BATCH_SIZE
print('steps_per_epoch', steps_per_epoch)
kl.train(train_gen,
val_gen,
saved_model_path=model_path,
steps=steps_per_epoch,
train_split=cfg.TRAIN_TEST_SPLIT)
def train(cfg, tub_names, model_name):
'''
use the specified data in tub_names to train an artifical neural network
saves the output trained model as model_name
'''
X_keys = ['cam/image_array', 'imu/imu_vec']
y_keys = ['user/angle', 'user/throttle']
def rt(record):
#record['user/angle'] = dk.utils.linear_bin(record['user/angle']) # Comment out if angle is not binned
record['imu/imu_vec'] = np.array(
ast.literal_eval(record['imu/imu_vec'].split('/')[-1]))
return record
#kl = KerasCategorical()
kl = KerasIMU()
print('tub_names', tub_names)
if not tub_names:
tub_names = os.path.join(cfg.DATA_PATH, '*')
tubgroup = TubGroup(tub_names)
train_gen, val_gen = tubgroup.get_train_val_gen(
X_keys,
y_keys,
record_transform=rt,
batch_size=cfg.BATCH_SIZE,
train_frac=cfg.TRAIN_TEST_SPLIT)
model_path = os.path.expanduser(model_name)
total_records = len(tubgroup.df)
total_train = int(total_records * cfg.TRAIN_TEST_SPLIT)
total_val = total_records - total_train
print('train: %d, validation: %d' % (total_train, total_val))
steps_per_epoch = total_train // cfg.BATCH_SIZE
print('steps_per_epoch', steps_per_epoch)
kl.train(train_gen,
val_gen,
saved_model_path=model_path,
steps=steps_per_epoch,
train_split=cfg.TRAIN_TEST_SPLIT)
def test_KerasIMU():
k = KerasIMU(num_outputs=2, num_imu_inputs=7)
assert k.model is not None
def drive(cfg, model_path=None, use_joystick=False):
'''
Construct a working robotic vehicle from many parts.
Each part runs as a job in the Vehicle loop, calling either
it's run or run_threaded method depending on the constructor flag `threaded`.
All parts are updated one after another at the framerate given in
cfg.DRIVE_LOOP_HZ assuming each part finishes processing in a timely manner.
Parts may have named outputs and inputs. The framework handles passing named outputs
to parts requesting the same named input.
'''
#Initialize car
V = dk.vehicle.Vehicle()
cam = PiCamera(resolution=cfg.CAMERA_RESOLUTION)
V.add(cam, outputs=['cam/image_array'], threaded=True)
imu = Mpu6050()
V.add(imu, outputs=['imu/imu_vec'], threaded=True)
# Lidar
imu = B0602Lidar()
V.add(imu, outputs=['lidar/lidar_measurements'], threaded=True)
if use_joystick or cfg.USE_JOYSTICK_AS_DEFAULT:
#modify max_throttle closer to 1.0 to have more power
#modify steering_scale lower than 1.0 to have less responsive steering
ctr = JoystickController(
max_throttle=cfg.JOYSTICK_MAX_THROTTLE,
steering_scale=cfg.JOYSTICK_STEERING_SCALE,
auto_record_on_throttle=cfg.AUTO_RECORD_ON_THROTTLE)
else:
#This web controller will create a web server that is capable
#of managing steering, throttle, and modes, and more.
ctr = LocalWebController()
V.add(ctr,
inputs=['cam/image_array'],
outputs=['user/angle', 'user/throttle', 'user/mode', 'recording'],
threaded=True)
#See if we should even run the pilot module.
#This is only needed because the part run_condition only accepts boolean
def pilot_condition(mode):
if mode == 'user':
return False
else:
return True
pilot_condition_part = Lambda(pilot_condition)
V.add(pilot_condition_part, inputs=['user/mode'], outputs=['run_pilot'])
# Add CV Here
#img_threshold = ImgThreshold()
#V.add(img_threshold, inputs=['cam/image_array'], outputs=['cam/image_array'])
#img_grayscale = ImgGreyscale()
#V.add(img_grayscale, inputs=['cam/image_array'], outputs=['cam/image_array'])
#img_canny = ImgCanny()
#V.add(img_canny, inputs=['cam/image_array'], outputs=['cam/image_array'])
# Make sure image size is correct
#xsize = 160
#ysize = 120
#vertices = np.array([[(0,48),(xsize,48),(xsize,ysize),(0,ysize)]], dtype=np.int32)
#img_mask = ImgMask(vertices)
#V.add(img_mask, inputs=['cam/image_array'], outputs=['cam/image_array'])
#img_stack = ImgStack()
#V.add(img_stack, inputs=['cam/image_array'], outputs=['cam/image_array'])
#Run the pilot if the mode is not user.
#kl = KerasCategorical()
kl = KerasIMU()
if model_path:
kl.load(model_path)
V.add(kl,
inputs=['cam/image_array', 'imu/imu_vec'],
outputs=['pilot/angle', 'pilot/throttle'],
run_condition='run_pilot')
#Choose what inputs should change the car.
def drive_mode(mode, user_angle, user_throttle, pilot_angle,
pilot_throttle):
if mode == 'user':
return user_angle, user_throttle
elif mode == 'local_angle':
return pilot_angle, user_throttle
else:
return pilot_angle, pilot_throttle
drive_mode_part = Lambda(drive_mode)
V.add(drive_mode_part,
inputs=[
'user/mode', 'user/angle', 'user/throttle', 'pilot/angle',
'pilot/throttle'
],
outputs=['angle', 'throttle'])
steering_controller = PCA9685(cfg.STEERING_CHANNEL)
steering = PWMSteering(controller=steering_controller,
left_pulse=cfg.STEERING_LEFT_PWM,
right_pulse=cfg.STEERING_RIGHT_PWM)
throttle_controller = PCA9685(cfg.THROTTLE_CHANNEL)
throttle = PWMThrottle(controller=throttle_controller,
max_pulse=cfg.THROTTLE_FORWARD_PWM,
zero_pulse=cfg.THROTTLE_STOPPED_PWM,
min_pulse=cfg.THROTTLE_REVERSE_PWM)
V.add(steering, inputs=['angle'])
V.add(throttle, inputs=['throttle'])
#add tub to save data
inputs = [
'cam/image_array', 'user/angle', 'user/throttle', 'user/mode',
'imu/imu_vec'
]
types = ['image_array', 'float', 'float', 'str', 'list']
th = TubHandler(path=cfg.DATA_PATH)
tub = th.new_tub_writer(inputs=inputs, types=types)
V.add(tub, inputs=inputs, run_condition='recording')
#run the vehicle
V.start(rate_hz=cfg.DRIVE_LOOP_HZ, max_loop_count=cfg.MAX_LOOPS)
print("You can now go to <your pi ip address>:8887 to drive your car.")
def drive(cfg,
model_path=None,
use_joystick=False,
model_type=None,
camera_type='single'):
'''
Construct a working robotic vehicle from many parts.
Each part runs as a job in the Vehicle loop, calling either
it's run or run_threaded method depending on the constructor flag `threaded`.
All parts are updated one after another at the framerate given in
cfg.DRIVE_LOOP_HZ assuming each part finishes processing in a timely manner.
Parts may have named outputs and inputs. The framework handles passing named outputs
to parts requesting the same named input.
'''
if model_type is None:
model_type = "categorical"
stereo_cam = camera_type == "stereo"
#Initialize car
V = dk.vehicle.Vehicle()
if stereo_cam:
from donkeycar.parts.camera import Webcam
camA = Webcam(image_w=cfg.IMAGE_W, image_h=cfg.IMAGE_H, iCam=0)
V.add(camA, outputs=['cam/image_array_a'], threaded=True)
camB = Webcam(image_w=cfg.IMAGE_W, image_h=cfg.IMAGE_H, iCam=1)
V.add(camB, outputs=['cam/image_array_b'], threaded=True)
def stereo_pair(image_a, image_b):
if image_a is not None and image_b is not None:
width, height, _ = image_a.shape
grey_a = dk.utils.rgb2gray(image_a)
grey_b = dk.utils.rgb2gray(image_b)
# Added by Felix
depth = capture.stereo_depth(image_a, image_b, leftMapX,
leftMapY, rightMapX, rightMapY,
leftROI, rightROI, imageSize)
stereo_image = np.zeros([width, height, 3],
dtype=np.dtype('B'))
stereo_image[..., 0] = np.reshape(grey_a, (width, height))
stereo_image[..., 1] = np.reshape(grey_b, (width, height))
stereo_image[..., 2] = np.reshape(depth, (width, height))
# --------------
else:
stereo_image = []
return np.array(stereo_image)
image_sterero_pair_part = Lambda(stereo_pair)
V.add(image_sterero_pair_part,
inputs=['cam/image_array_a', 'cam/image_array_b'],
outputs=['cam/image_array'])
else:
print("cfg.CAMERA_TYPE", cfg.CAMERA_TYPE)
if cfg.CAMERA_TYPE == "PICAM":
from donkeycar.parts.camera import PiCamera
cam = PiCamera(image_w=cfg.IMAGE_W, image_h=cfg.IMAGE_H)
elif cfg.CAMERA_TYPE == "WEBCAM":
from donkeycar.parts.camera import Webcam
cam = Webcam(image_w=cfg.IMAGE_W, image_h=cfg.IMAGE_H)
V.add(cam, outputs=['cam/image_array'], threaded=True)
if use_joystick or cfg.USE_JOYSTICK_AS_DEFAULT:
#modify max_throttle closer to 1.0 to have more power
#modify steering_scale lower than 1.0 to have less responsive steering
ctr = JoystickController(
throttle_scale=cfg.JOYSTICK_MAX_THROTTLE,
steering_scale=cfg.JOYSTICK_STEERING_SCALE,
auto_record_on_throttle=cfg.AUTO_RECORD_ON_THROTTLE)
else:
#This web controller will create a web server that is capable
#of managing steering, throttle, and modes, and more.
ctr = LocalWebController()
V.add(ctr,
inputs=['cam/image_array'],
outputs=['user/angle', 'user/throttle', 'user/mode', 'recording'],
threaded=True)
#this throttle filter will allow one tap back for esc reverse
th_filter = ThrottleFilter()
V.add(th_filter, inputs=['user/throttle'], outputs=['user/throttle'])
#See if we should even run the pilot module.
#This is only needed because the part run_condition only accepts boolean
def pilot_condition(mode):
if mode == 'user':
return False
else:
return True
pilot_condition_part = Lambda(pilot_condition)
V.add(pilot_condition_part, inputs=['user/mode'], outputs=['run_pilot'])
def led_cond(mode, recording, num_records, behavior_state):
'''
returns a blink rate. 0 for off. -1 for on. positive for rate.
'''
if num_records is not None and num_records % 10 == 0:
print("recorded", num_records, "records")
if behavior_state is not None and model_type == 'behavior':
r, g, b = cfg.BEHAVIOR_LED_COLORS[behavior_state]
led.set_rgb(r, g, b)
return -1 #solid on
if recording:
return -1 #solid on
elif mode == 'user':
return 1
elif mode == 'local_angle':
return 0.5
elif mode == 'local':
return 0.1
return 0
led_cond_part = Lambda(led_cond)
V.add(
led_cond_part,
inputs=['user/mode', 'recording', "tub/num_records", 'behavior/state'],
outputs=['led/blink_rate'])
if cfg.HAVE_RGB_LED:
from donkeycar.parts.led_status import RGB_LED
led = RGB_LED(cfg.LED_PIN_R, cfg.LED_PIN_G, cfg.LED_PIN_B,
cfg.LED_INVERT)
led.set_rgb(cfg.LED_R, cfg.LED_G, cfg.LED_B)
V.add(led, inputs=['led/blink_rate'])
#IMU
if cfg.HAVE_IMU:
imu = Mpu6050()
V.add(imu,
outputs=[
'imu/acl_x', 'imu/acl_y', 'imu/acl_z', 'imu/gyr_x',
'imu/gyr_y', 'imu/gyr_z'
],
threaded=True)
#Behavioral state
if model_type == "behavior":
bh = BehaviorPart(cfg.BEHAVIOR_LIST)
V.add(bh,
outputs=[
'behavior/state', 'behavior/label',
"behavior/one_hot_state_array"
])
try:
ctr.set_button_down_trigger('L1', bh.increment_state)
except:
pass
kl = KerasBehavioral(num_outputs=2,
num_behavior_inputs=len(cfg.BEHAVIOR_LIST))
inputs = ['cam/image_array', "behavior/one_hot_state_array"]
#IMU
elif model_type == "imu":
assert (cfg.HAVE_IMU)
#Run the pilot if the mode is not user.
kl = KerasIMU(num_outputs=2, num_imu_inputs=6)
inputs = [
'cam/image_array', 'imu/acl_x', 'imu/acl_y', 'imu/acl_z',
'imu/gyr_x', 'imu/gyr_y', 'imu/gyr_z'
]
else:
if model_type == "linear":
kl = KerasLinear()
elif model_type == "3d":
kl = Keras3D_CNN(seq_length=cfg.SEQUENCE_LENGTH)
elif model_type == "rnn":
kl = KerasRNN_LSTM(seq_length=cfg.SEQUENCE_LENGTH)
else:
kl = KerasCategorical()
inputs = ['cam/image_array']
if model_path:
kl.load(model_path)
V.add(kl,
inputs=inputs,
outputs=['pilot/angle', 'pilot/throttle'],
run_condition='run_pilot')
#Choose what inputs should change the car.
def drive_mode(mode, user_angle, user_throttle, pilot_angle,
pilot_throttle):
if mode == 'user':
return user_angle, user_throttle
elif mode == 'local_angle':
return pilot_angle, user_throttle
else:
return pilot_angle, pilot_throttle
drive_mode_part = Lambda(drive_mode)
V.add(drive_mode_part,
inputs=[
'user/mode', 'user/angle', 'user/throttle', 'pilot/angle',
'pilot/throttle'
],
outputs=['angle', 'throttle'])
steering_controller = PCA9685(cfg.STEERING_CHANNEL,
cfg.PCA9685_I2C_ADDR,
busnum=cfg.PCA9685_I2C_BUSNUM)
steering = PWMSteering(controller=steering_controller,
left_pulse=cfg.STEERING_LEFT_PWM,
right_pulse=cfg.STEERING_RIGHT_PWM)
throttle_controller = PCA9685(cfg.THROTTLE_CHANNEL,
cfg.PCA9685_I2C_ADDR,
busnum=cfg.PCA9685_I2C_BUSNUM)
throttle = PWMThrottle(controller=throttle_controller,
max_pulse=cfg.THROTTLE_FORWARD_PWM,
zero_pulse=cfg.THROTTLE_STOPPED_PWM,
min_pulse=cfg.THROTTLE_REVERSE_PWM)
V.add(steering, inputs=['angle'])
V.add(throttle, inputs=['throttle'])
#add tub to save data
inputs = ['cam/image_array', 'user/angle', 'user/throttle', 'user/mode']
types = ['image_array', 'float', 'float', 'str']
if cfg.TRAIN_BEHAVIORS:
inputs += [
'behavior/state', 'behavior/label', "behavior/one_hot_state_array"
]
types += ['int', 'str', 'vector']
if cfg.HAVE_IMU:
inputs += [
'imu/acl_x', 'imu/acl_y', 'imu/acl_z', 'imu/gyr_x', 'imu/gyr_y',
'imu/gyr_z'
]
types += ['float', 'float', 'float', 'float', 'float', 'float']
th = TubHandler(path=cfg.DATA_PATH)
tub = th.new_tub_writer(inputs=inputs, types=types)
V.add(tub,
inputs=inputs,
outputs=["tub/num_records"],
run_condition='recording')
if type(ctr) is LocalWebController:
print("You can now go to <your pi ip address>:8887 to drive your car.")
elif type(ctr) is JoystickController:
print("You can now move your joystick to drive your car.")
#tell the controller about the tub
ctr.set_tub(tub)
#run the vehicle for 20 seconds
V.start(rate_hz=cfg.DRIVE_LOOP_HZ, max_loop_count=cfg.MAX_LOOPS)
def get_model_by_type(model_type: str, cfg: 'Config') -> 'KerasPilot':
'''
given the string model_type and the configuration settings in cfg
create a Keras model and return it.
'''
from donkeycar.parts.keras import KerasCategorical, KerasLinear, \
KerasInferred, KerasIMU, KerasMemory, KerasBehavioral, KerasLocalizer, \
KerasLSTM, Keras3D_CNN
from donkeycar.parts.interpreter import KerasInterpreter, TfLite, TensorRT
if model_type is None:
model_type = cfg.DEFAULT_MODEL_TYPE
logger.info(f'get_model_by_type: model type is: {model_type}')
input_shape = (cfg.IMAGE_H, cfg.IMAGE_W, cfg.IMAGE_DEPTH)
if 'tflite_' in model_type:
interpreter = TfLite()
used_model_type = model_type.replace('tflite_', '')
elif 'tensorrt_' in model_type:
interpreter = TensorRT()
used_model_type = model_type.replace('tensorrt_', '')
else:
interpreter = KerasInterpreter()
used_model_type = model_type
used_model_type = EqMemorizedString(used_model_type)
if used_model_type == "linear":
kl = KerasLinear(interpreter=interpreter, input_shape=input_shape)
elif used_model_type == "categorical":
kl = KerasCategorical(
interpreter=interpreter,
input_shape=input_shape,
throttle_range=cfg.MODEL_CATEGORICAL_MAX_THROTTLE_RANGE)
elif used_model_type == 'inferred':
kl = KerasInferred(interpreter=interpreter, input_shape=input_shape)
elif used_model_type == "imu":
kl = KerasIMU(interpreter=interpreter, input_shape=input_shape)
elif used_model_type == "memory":
mem_length = getattr(cfg, 'SEQUENCE_LENGTH', 3)
mem_depth = getattr(cfg, 'MEM_DEPTH', 0)
kl = KerasMemory(interpreter=interpreter,
input_shape=input_shape,
mem_length=mem_length,
mem_depth=mem_depth)
elif used_model_type == "behavior":
kl = KerasBehavioral(
interpreter=interpreter,
input_shape=input_shape,
throttle_range=cfg.MODEL_CATEGORICAL_MAX_THROTTLE_RANGE,
num_behavior_inputs=len(cfg.BEHAVIOR_LIST))
elif used_model_type == 'localizer':
kl = KerasLocalizer(interpreter=interpreter,
input_shape=input_shape,
num_locations=cfg.NUM_LOCATIONS)
elif used_model_type == 'rnn':
kl = KerasLSTM(interpreter=interpreter,
input_shape=input_shape,
seq_length=cfg.SEQUENCE_LENGTH)
elif used_model_type == '3d':
kl = Keras3D_CNN(interpreter=interpreter,
input_shape=input_shape,
seq_length=cfg.SEQUENCE_LENGTH)
else:
known = [
k + u for k in ('', 'tflite_', 'tensorrt_')
for u in used_model_type.mem
]
raise ValueError(
f"Unknown model type {model_type}, supported types are"
f" { ', '.join(known)}")
return kl