dctian

def drive(self, speed=__INITIAL_SPEED):
    self.back_wheels.speed = speed
    while self.camera.isOpened():
        _, image_lane = self.camera.read()
        image_objs = image_lane.copy()

        image_objs = self.process_objects_on_road(image_objs)

        image_lane = self.follow_lane(image_lane)

dctian

class ObjectsOnRoadProcessor(object):
    """
    This class 1) detects what objects (namely traffic signs and people) are on the road
    and 2) controls the car navigation (speed/steering) accordingly
    """

    def __init__(self,
                 car=None,
                 speed_limit=40,
                 model='/home/pi/DeepPiCar/models/object_detection/data/model_result/road_signs_quantized_edgetpu.tflite',

dctian

class SpeedLimit(TrafficObject):

    def __init__(self, speed_limit):
        self.speed_limit = speed_limit

    def set_car_state(self, car_state):
        logging.debug('speed limit: set limit to %d' % self.speed_limit)
        car_state['speed_limit'] = self.speed_limit

dctian

class RedTrafficLight(TrafficObject):
    def set_car_state(self, car_state):
        logging.debug('red light: stopping car')
        car_state['speed'] = 0

class Pedestrian(TrafficObject):
    def set_car_state(self, car_state):
        logging.debug('pedestrian: stopping car')
        car_state['speed'] = 0

dctian

class TrafficObject(object):

    def set_car_state(self, car_state):
        pass

    @staticmethod
    def is_close_by(obj, frame_height, min_height_pct=0.05):
        # default: if a sign is 10% of the height of frame
        obj_height = obj.bounding_box[1][1]-obj.bounding_box[0][1]
        return obj_height / frame_height > min_height_pct

dctian

# creates the frozen inference graph in fine_tune_model
# there is an "Incomplete shape" message.  but we can safely ignore that. 
!python /content/models/research/object_detection/export_inference_graph.py \
    --input_type=image_tensor \
    --pipeline_config_path={pipeline_fname} \
    --output_directory='{output_directory}' \
    --trained_checkpoint_prefix='{last_model_path}'

# create the tensorflow lite graph
!python /content/models/research/object_detection/export_tflite_ssd_graph.py \

dctian

num_steps = 2000
num_eval_steps = 50
model_dir = '/content/gdrive/My Drive/Colab Notebooks/TransferLearning/Training'
pipeline_file = 'ssd_mobilenet_v2_quantized_300x300_coco.config'

!python /content/models/research/object_detection/model_main.py \
    --pipeline_config_path={pipeline_fname} \
    --model_dir='{model_dir}' \
    --alsologtostderr \
    --num_train_steps={num_steps} \

dctian

MODEL_FILE = MODEL + '.tar.gz'
DOWNLOAD_BASE = 'http://download.tensorflow.org/models/object_detection/'
DEST_DIR = '/content/models/research/pretrained_model'

if not (os.path.exists(MODEL_FILE)):
    urllib.request.urlretrieve(DOWNLOAD_BASE + MODEL_FILE, MODEL_FILE)

tar = tarfile.open(MODEL_FILE)
tar.extractall()
tar.close()

dctian

repo_dir_path = '/content/DeepPiCar'
%cd {repo_dir_path}/models/object_detection

# Convert train folder annotation xml files to a single csv file,
# generate the `label_map.pbtxt` file to `data/` directory as well.
!python code/xml_to_csv.py -i data/images/train -o data/annotations/train_labels.csv -l data/annotations

# Convert test folder annotation xml files to a single csv.
!python code/xml_to_csv.py -i data/images/test -o data/annotations/test_labels.csv

dctian

# If you forked the repository, you can replace the link.
repo_url = 'https://github.com/dctian/DeepPiCar'

# Number of training steps.
num_steps = 1000  # 200000

# Number of evaluation steps.
num_eval_steps = 50

# model configs are from Model Zoo github: