def __init__(self, **kwargs):
super().__init__(**kwargs)
self.pattern_name = "Music"
CONFIGS['n_pixels'] = self.strip_length
# init visualizer
self.vis = Visualizer(CONFIGS)
# dict used to set the visualizer effect
self.effect_dict = dict(
spectrum=self.vis.visualize_spectrum,
energy=self.vis.visualize_energy,
scroll=self.vis.visualize_scroll,
)
# name of the effect to be used
self.effect = 'spectrum'
self.modifiers = dict(visualizer=self.effect, )
# attributes for the mic
self.p = None
self.stream = None
self.frames_per_buffer = int(CONFIGS['mic_rate'] / CONFIGS['fps'])
self.setup()
def __init__(self, config, host='192.168.137.183', port=8000):
super(ClientThread, self).__init__()
self.config = config
self.data = []
self.host = host
self.port = port
self.path = config.root + config.test_path + 'data' + self.host + 'temp.pkl'
self.device = torch.device("cpu")
parser = argparse.ArgumentParser(
description='Wifi Indoor Positioning System')
args = parser.parse_args()
args.feat_dim = config.n_address
args.dropout = 0.0
self.model = DNN_8(args)
self.model.load_state_dict(
torch.load(
"./checkpoints/address_128_8_layer_jitter_0.05_valsplit_0.01/models/model.t7",
map_location=self.device))
self.model.to(self.device)
self.model.eval()
print(self.model)
self.do_run = True
self.data = []
self.v = Visualizer()
self.x = None
self.y = None
self.alpha = 0.5
def upload_to_kibana(iters):
'''
setup connection to kibana and upload data
:param iters: tweets in each partition
:return: None
'''
res = list(iters)
if not len(res): return
Visualizer.upload(res)
def __init__(self, actor, critic, episode_num, sim_world,
visualization_speed):
self.actor = actor
self.critic = critic
self.episode_num = episode_num
self.sim_world = sim_world
self.visualizer = Visualizer(self.sim_world.get_board(),
self.sim_world.get_player(),
visualization_speed)
def __init__(self):
self.visualizer = Visualizer()
self.render = False
self.food = []
self.poison = []
self.generation = 1
self.bounds = WORLD_BOUNDS # tuple e.g. (x, y)
self.agent = Agent()
#self.pool = None if NUM_CORES < 2 else multiprocessing.Pool(NUM_CORES)
self.init()
def __init__(self, config, host='192.168.137.183', port=8000):
super(ClientThread, self).__init__()
self.config = config
self.data = []
self.host = host
self.port = port
self.path = config.root + config.test_path + 'data' + self.host + 'temp.pkl'
self.device = torch.device("cpu")
self.model = torch.load(config.root + config.model_path,
map_location=torch.device('cpu')).to(
self.device)
self.model.eval()
self.do_run = True
self.data = []
self.v = Visualizer()
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--filename',
type=str,
default='./visualization/data/itr_2322.pkl'
) # defaultIS.h5/snapshots/iter0000480
parser.add_argument('--vid',
type=str,
default='./visualization/videos/contworldvid.mp4')
parser.add_argument('--deterministic', action='store_true', default=False)
parser.add_argument('--heuristic', action='store_true', default=False)
parser.add_argument('--evaluate', action='store_true', default=False)
parser.add_argument('--n_trajs', type=int, default=48)
parser.add_argument('--n_steps', type=int, default=1000)
parser.add_argument('--same_con_pol', action='store_true', default=False)
args = parser.parse_args()
with open('./visualization/data/params.json', 'r') as df:
train_args = json.load(df)
env = MAContWorld(
n_rovers=train_args['n_rovers'],
n_areas_of_int=train_args['n_areas_of_int'],
n_coop=train_args['n_coop'],
n_crater=train_args['n_crater'],
n_sensors=train_args['n_sensors'],
scout_reward=train_args['scout_reward'],
crater_reward=train_args['crater_reward'],
control_penalty=-.5,
reward_mech='Local',
encounter_reward=train_args[
'encounter_reward'], #train_args['encounter_reward'],
n_obstacles=3,
addid=True,
_speed_features=True,
obstacle_loc=None)
env.reset()
if train_args['buffer_size'] > 1:
env = ObservationBuffer(env, train_args['buffer_size'])
hpolicy = None # heuristic policy
if args.evaluate:
minion = Evaluator(env, train_args, args.n_steps, args.n_trajs,
args.deterministic,
'heuristic' if args.heuristic else 'rllab')
evr = minion(args.filename,
same_con_pol=args.same_con_pol,
hpolicy=hpolicy)
from tabulate import tabulate
print(tabulate(evr, headers='keys'))
else:
minion = Visualizer(env, train_args, args.n_steps, args.n_trajs,
args.deterministic,
'heuristic' if args.heuristic else 'rllab')
rew, info = minion(args.filename, vid=args.vid, hpolicy=hpolicy)
pprint.pprint(rew)
pprint.pprint(info)
def visualize(self, knn_accuracy, knn_k, kmeans_accuracy, kmeans_k):
visualize = Visualizer()
if knn_accuracy != [] and kmeans_accuracy != []:
visualize.kmeans_knn(knn_accuracy, knn_k, kmeans_accuracy,
kmeans_k)
elif kmeans_accuracy != []:
visualize.kmeans(kmeans_accuracy, kmeans_k)
elif knn_accuracy != []:
visualize.knn(knn_accuracy, knn_k)
def visualize(self, knn_accuracy, knn_k, kmeans_accuracy, kmeans_k):
visualize = Visualizer()
if knn_accuracy != [] and kmeans_accuracy != []:
visualize.kmeans_knn(knn_accuracy, knn_k, kmeans_accuracy, kmeans_k)
elif kmeans_accuracy != []:
visualize.kmeans(kmeans_accuracy, kmeans_k)
elif knn_accuracy != []:
visualize.knn(knn_accuracy, knn_k)
class Graph:
def __init__(self):
self.graph = defaultdict(list) # default_factory() of type list
self.v = Visualizer()
self.result = ''
def add_edge(self, u, v):
self.graph[u].append(v)
self.graph[v].append(u) # Undirected graph is bidrectional
self.v.add_edge(u, v)
def visualize(self):
self.v.visualize()
def print(self, starting_node):
print("DFS from vertex {} = {}".format(starting_node, self.result))
# The function to do DFS traversal. It uses
# recursive DFSUtil()
def DFS(self, v):
'''
6 starting vertices from 0 to 5
=> bool array with 6 elements
'''
visited_arr = [False] * (max(self.graph) + 1)
'''
recursive function that takes
the index of node and
visited array
'''
self.DFSUtil(v, visited_arr)
# Recursive function for DFS with Backtracking
def DFSUtil(self, v, visited):
visited[v] = True
# Adding V to result
self.result += str(v) + " "
# Traversing vertices adjacent to this vertex
for i in self.graph[v]:
if visited[i] == False:
self.DFSUtil(i, visited)
class SegmentPlotCallback(keras.callbacks.Callback):
def __init__(self, configuration, data_loader):
self.configuration = configuration
self.visualization_parent_path = configuration.VISUALIZATION_PARENT_PATH
self.visualizer = Visualizer(configuration)
self.data_loader = data_loader
self.train_path = os.path.join(self.visualization_parent_path,
'train_sample_across_epochs')
self.test_path = os.path.join(self.visualization_parent_path,
'test_sample_across_epochs')
self.validation_path = os.path.join(self.visualization_parent_path,
'validation_sample_across_epochs')
# change the following numbers to choose other segments for visualization
self.train_segment_index = configuration.TRAIN_SEGMENT_INDEX
self.test_segment_index = configuration.TEST_SEGMENT_INDEX
self.validation_segment_index = configuration.VALIDATION_SEGMENT_INDEX
#
if not os.path.exists(self.train_path):
os.makedirs(self.train_path)
if not os.path.exists(self.test_path):
os.makedirs(self.test_path)
if not os.path.exists(self.validation_path):
os.makedirs(self.validation_path)
def on_epoch_end(self, epoch, logs={}):
self.visualizer.visualize_prediction_segments(
self.model,
np.expand_dims(
self.data_loader.train_segments[self.train_segment_index], 0),
path=self.train_path,
initial_segment=self.train_segment_index,
epoch_id='epoch_' + str(epoch),
layer_filter_list=['mclnn'],
first_mclnn_only=True)
self.visualizer.visualize_prediction_segments(
self.model,
np.expand_dims(
self.data_loader.test_segments[self.test_segment_index], 0),
path=self.test_path,
initial_segment=self.test_segment_index,
epoch_id='epoch_' + str(epoch),
layer_filter_list=['mclnn'],
first_mclnn_only=True)
self.visualizer.visualize_prediction_segments(
self.model,
np.expand_dims(
self.data_loader.validation_segments[
self.validation_segment_index], 0),
path=self.validation_path,
initial_segment=self.validation_segment_index,
epoch_id='epoch_' + str(epoch),
layer_filter_list=['mclnn'],
first_mclnn_only=True)
class Graph:
def __init__(self):
self.graph = defaultdict(list) # default_factory() of type list
self.v = Visualizer()
self.result = ""
def add_edge(self, u, v):
self.graph[u].append(v)
self.graph[v].append(u) # Undirected graph is bidrectional
self.v.add_edge(u, v)
# No BFS util required, no recusion required
# Will be solved with just iteration
# Passing children of nodes iteratively and visit them
def BFS(self, starting_node):
visited_arr = [False] * (len(self.graph))
queue = list()
# enqueue it
queue.append(starting_node)
visited_arr[starting_node] = True
while queue:
node = queue.pop(0)
self.result += str(node) + " "
for i in self.graph[node]:
if visited_arr[i] == False:
queue.append(i)
visited_arr[i] = True
def print(self, starting_node):
print("BFS from vertex {} = {}".format(starting_node, self.result))
def visualize(self):
self.v.visualize()
def __init__(self, configuration, data_loader):
self.configuration = configuration
self.visualization_parent_path = configuration.VISUALIZATION_PARENT_PATH
self.visualizer = Visualizer(configuration)
self.data_loader = data_loader
self.train_path = os.path.join(self.visualization_parent_path,
'train_sample_across_epochs')
self.test_path = os.path.join(self.visualization_parent_path,
'test_sample_across_epochs')
self.validation_path = os.path.join(self.visualization_parent_path,
'validation_sample_across_epochs')
# change the following numbers to choose other segments for visualization
self.train_segment_index = configuration.TRAIN_SEGMENT_INDEX
self.test_segment_index = configuration.TEST_SEGMENT_INDEX
self.validation_segment_index = configuration.VALIDATION_SEGMENT_INDEX
#
if not os.path.exists(self.train_path):
os.makedirs(self.train_path)
if not os.path.exists(self.test_path):
os.makedirs(self.test_path)
if not os.path.exists(self.validation_path):
os.makedirs(self.validation_path)
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.pattern_name = "Music"
CONFIGS['n_pixels'] = self.strip_length
# init visualizer
self.vis = Visualizer(CONFIGS)
# dict used to set the visualizer effect
self.effect_dict = dict(
spectrum=self.vis.visualize_spectrum,
energy=self.vis.visualize_energy,
scroll=self.vis.visualize_scroll,
)
# name of the effect to be used
self.effect = Modifier('visualizer', "spectrum", options=list(self.effect_dict.keys()),
on_change=self.on_change)
self._rate = Modifier('Delay', 0, minimum=0, maximum=0)
self.modifiers = dict(
effect=self.effect,
)
# attributes for the mic
self.p = None
self.stream = None
self.frames_per_buffer = int(CONFIGS['mic_rate'] / CONFIGS['fps'])
try:
# do not initialize if the pattern is temp
if kwargs['handler'] is not None:
self.setup()
except OSError:
music_logger.warning(f"Could not initialize the audio stream")
class Music(Default):
"""
Music reactive leds
"""
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.pattern_name = "Music"
CONFIGS['n_pixels'] = self.strip_length
# init visualizer
self.vis = Visualizer(CONFIGS)
# dict used to set the visualizer effect
self.effect_dict = dict(
spectrum=self.vis.visualize_spectrum,
energy=self.vis.visualize_energy,
scroll=self.vis.visualize_scroll,
)
# name of the effect to be used
self.effect = Modifier('visualizer', "spectrum", options=list(self.effect_dict.keys()),
on_change=self.on_change)
self._rate = Modifier('Delay', 0, minimum=0, maximum=0)
self.modifiers = dict(
effect=self.effect,
)
# attributes for the mic
self.p = None
self.stream = None
self.frames_per_buffer = int(CONFIGS['mic_rate'] / CONFIGS['fps'])
try:
# do not initialize if the pattern is temp
if kwargs['handler'] is not None:
self.setup()
except OSError:
music_logger.warning(f"Could not initialize the audio stream")
def setup(self):
"""
Setup stream
"""
self.p = pyaudio.PyAudio()
self.stream = self.p.open(format=pyaudio.paInt16,
channels=1,
rate=CONFIGS['mic_rate'],
input=True,
frames_per_buffer=self.frames_per_buffer)
music_logger.info(f"Audio stream initialized with {CONFIGS['fps']} fps")
def on_change(self, value):
"""
Set the effect to a certain value and change the visualization effect in the vis calss
"""
try:
ef = self.effect_dict[value]
self.vis.visualization_effect = ef
except KeyError as e:
music_logger.warning(f"Error for key {value}\n{e}")
@property
def rate(self):
"""
Rate should always be zero here
"""
return self._rate
@rate.setter
def rate(self, value):
"""
Cannot change the value of rate since music must be real time
:param value:
:return:
"""
pass
def read_audio(self):
"""
Read audio and return it
"""
try:
y = np.fromstring(self.stream.read(self.frames_per_buffer, exception_on_overflow=False),
dtype=np.int16)
y = y.astype(np.float32)
return y
except IOError:
music_logger.warning('Audio buffer has overflowed')
except AttributeError:
music_logger.error("Could not read from audio buffer, do you have a microphone?")
self.close()
def fill(self):
"""
Read from audio stream and set pixels
"""
# read audio input, can also be none when the mic has not started yet
output = self.read_audio()
try:
# use visualization
pixels, _ = self.vis.audio_to_rgb(output)
# Truncate values and cast to integer
pixels = np.clip(pixels, 0, 255).astype(int)
# Optional gamma correction
pixels = _gamma[pixels]
r, g, b = pixels
for idx in range(len(r)):
self.pixels[idx]['color'] = (r[idx], g[idx], b[idx], 255)
except TypeError:
pass
def close(self):
"""
Call super method and close audio stream
"""
super(Music, self).close()
try:
# try to stop the stream if any
self.stream.stop_stream()
self.stream.close()
except AttributeError:
# if there is no recognized microphone then the close operation will fail on the stream
pass
finally:
# terminate the py audio and log
self.p.terminate()
music_logger.info("Audio stream stopped")
parser.add_argument('--tail', type=str, default='', help='specific name')
parser.add_argument('--visualize',
action='store_true',
default=False,
help='Visualize the loss curve in visdom windows')
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
print(args)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if args.visualize:
vis = Visualizer(
env='{}_{}_{}'.format(args.dataset, args.encoder, args.tail))
# ================== Model & Log Save Folder =================== #
log = None
# Save model and meta-data. Always saves in a new folder.
if args.save_folder:
if args.restore:
pass
else:
exp_counter = 0
save_folder = os.path.join(
args.save_folder,
'{}_{}_{}_exp{}'.format(args.dataset, args.encoder, args.tail,
exp_counter))
while os.path.isdir(save_folder):
exp_counter += 1
def makeDiagram(n_quadrotors,
n_balls,
use_visualizer=False,
trajectory_u=None,
trajectory_x=None,
trajectory_K=None):
builder = DiagramBuilder()
# Setup quadrotor plants and controllers
quadrotor_plants = []
quadrotor_controllers = []
for i in range(n_quadrotors):
new_quad = Quadrotor2D(n_quadrotors=n_quadrotors - 1, n_balls=n_balls)
new_quad.set_name('quad_' + str(i))
plant = builder.AddSystem(new_quad)
quadrotor_plants.append(plant)
# Setup ball plants
ball_plants = []
for i in range(n_balls):
new_ball = Ball2D(n_quadrotors=n_quadrotors, n_balls=n_balls - 1)
new_ball.set_name('ball_' + str(i))
plant = builder.AddSystem(new_ball)
ball_plants.append(plant)
# Connect all plants so that each object (quadrotor or ball) has access to all other object states as inputs
for i in range(n_quadrotors):
for j in range(n_quadrotors):
if i == j:
continue
k = j if j < i else j - 1
builder.Connect(quadrotor_plants[j].get_output_port(0),
quadrotor_plants[i].GetInputPort('quad_' + str(k)))
for j in range(n_balls):
builder.Connect(ball_plants[j].get_output_port(0),
quadrotor_plants[i].GetInputPort('ball_' + str(j)))
for i in range(n_balls):
for j in range(n_quadrotors):
builder.Connect(quadrotor_plants[j].get_output_port(0),
ball_plants[i].GetInputPort('quad_' + str(j)))
for j in range(n_balls):
if i == j:
continue
k = j if j < i else j - 1
builder.Connect(ball_plants[j].get_output_port(0),
ball_plants[i].GetInputPort('ball_' + str(k)))
# Setup visualization
if use_visualizer:
visualizer = builder.AddSystem(
Visualizer(n_quadrotors=n_quadrotors, n_balls=n_balls))
visualizer.set_name('visualizer')
for i in range(n_quadrotors):
builder.Connect(quadrotor_plants[i].get_output_port(0),
visualizer.get_input_port(i))
for i in range(n_balls):
builder.Connect(ball_plants[i].get_output_port(0),
visualizer.get_input_port(n_quadrotors + i))
# Setup trajectory source
if trajectory_x is not None and trajectory_u is not None and trajectory_K is not None:
demulti_u = builder.AddSystem(Demultiplexer(2 * n_quadrotors, 2))
demulti_u.set_name('feedforward input')
demulti_x = builder.AddSystem(Demultiplexer(6 * n_quadrotors, 6))
demulti_x.set_name('reference trajectory')
demulti_K = builder.AddSystem(Demultiplexer(12 * n_quadrotors, 12))
demulti_K.set_name('time-varying K')
for i in range(n_quadrotors):
ltv_lqr = builder.AddSystem(LTVController(6, 2))
ltv_lqr.set_name('LTV LQR ' + str(i))
builder.Connect(demulti_x.get_output_port(i),
ltv_lqr.get_input_port(0))
builder.Connect(quadrotor_plants[i].get_output_port(0),
ltv_lqr.get_input_port(1))
builder.Connect(demulti_u.get_output_port(i),
ltv_lqr.get_input_port(2))
builder.Connect(demulti_K.get_output_port(i),
ltv_lqr.get_input_port(3))
builder.Connect(ltv_lqr.get_output_port(0),
quadrotor_plants[i].get_input_port(0))
source_u = builder.AddSystem(TrajectorySource(trajectory_u))
source_u.set_name('source feedforward input trajectory')
source_x = builder.AddSystem(TrajectorySource(trajectory_x))
source_x.set_name('source reference trajectory')
demulti_source_x = builder.AddSystem(
Demultiplexer([6 * n_quadrotors, 4 * n_balls]))
demulti_source_x.set_name('quad and ball trajectories')
source_K = builder.AddSystem(TrajectorySource(trajectory_K))
source_K.set_name('source time-varying K')
builder.Connect(source_u.get_output_port(0),
demulti_u.get_input_port(0))
builder.Connect(source_x.get_output_port(0),
demulti_source_x.get_input_port(0))
builder.Connect(demulti_source_x.get_output_port(0),
demulti_x.get_input_port(0))
builder.Connect(source_K.get_output_port(0),
demulti_K.get_input_port(0))
else:
demulti_u = builder.AddSystem(Demultiplexer(2 * n_quadrotors, 2))
demulti_u.set_name('quadrotor input')
for i in range(n_quadrotors):
builder.Connect(demulti_u.get_output_port(i),
quadrotor_plants[i].get_input_port(0))
builder.ExportInput(demulti_u.get_input_port(0))
diagram = builder.Build()
return diagram
# spin up ssl-vision data polling to update gamestate
vision.start_updating(VISION_LOOP_SLEEP)
if not VISION_ONLY:
# spin up comms to send commands to robots
home_comms.start_sending(COMMS_SEND_LOOP_SLEEP)
# home_comms.start_receiving(COMMS_RECEIVE_LOOP_SLEEP)
if CONTROL_BOTH_TEAMS:
away_comms.start_sending(COMMS_SEND_LOOP_SLEEP)
# away_comms.start_sending(COMMS_RECEIVE_LOOP_SLEEP)
refbox.start_updating()
# spin up strategy threads to control the robots
home_strategy.start_controlling(HOME_STRATEGY, CONTROL_LOOP_SLEEP)
if CONTROL_BOTH_TEAMS:
away_strategy.start_controlling(AWAY_STRATEGY, CONTROL_LOOP_SLEEP)
# initialize visualizer to show robots on screen
visualizer = Visualizer(gamestate, home_strategy, away_strategy)
# start the game - now everything should be going
gamestate.start_game(GAME_LOOP_SLEEP)
# Prepare to be interrupted by user
exit_signal_received = False
def exit_gracefully(signum, frame):
global exit_signal_received
if exit_signal_received:
return
else:
exit_signal_received = True
print('Exiting Everything')
# clean up all threads
vision.stop_updating()
def run():
# ======================================= Initialization ======================================= #
all_folds_target_label = np.asarray([])
all_folds_majority_vote_cm = np.zeros(
(Config.NB_CLASSES, Config.NB_CLASSES), dtype=np.int)
all_folds_majority_vote_label = np.asarray([])
all_folds_probability_vote_cm = np.zeros(
(Config.NB_CLASSES, Config.NB_CLASSES), dtype=np.int)
all_folds_probability_vote_label = np.asarray([])
segment_size = sum(Config.LAYERS_ORDER_LIST) * 2 + Config.EXTRA_FRAMES
print('Segment without middle frame:' + str(segment_size))
is_visualization_called_flag = False # visualization is done for first fold only using this variable
# list of paths to the n-fold indices of the Training/Testing/Validation splits
# number of paths should be e.g. 30 for 3x10, where 3 is for the splits and 10 for the 10-folds
# Every 3 files are for one run to train and validate on 9-folds and test on the remaining fold.
folds_index_file_list = glob.glob(
os.path.join(Config.INDEX_PATH, "Fold*.hdf5"))
if len(folds_index_file_list) == 0:
print('Index path is not found = ' + Config.INDEX_PATH)
return
folds_index_file_list.sort()
cross_val_index_list = np.arange(
0, Config.SPLIT_COUNT * Config.CROSS_VALIDATION_FOLDS_COUNT,
Config.SPLIT_COUNT)
# ======================================= Start cross-validation ======================================= #
for j in range(cross_val_index_list[Config.INITIAL_FOLD_ID],
len(folds_index_file_list), Config.SPLIT_COUNT):
test_index_path = folds_index_file_list[j] if folds_index_file_list[
j].lower().endswith('_test.hdf5') else None
train_index_path = folds_index_file_list[
j + 1] if folds_index_file_list[j + 1].lower().endswith(
'_train.hdf5') else None
validation_index_path = folds_index_file_list[
j + 2] if folds_index_file_list[j + 2].lower().endswith(
'_validation.hdf5') else None
if None in [test_index_path, train_index_path, validation_index_path]:
print(
'Train / Validation / Test indices are not correctly assigned')
exit(1)
np.random.seed(0) # for reproducibility
data_loader = DataLoader()
mclnn_trainer = MCLNNTrainer()
# --------------------------------- Load data ----------------------------- #
data_loader.load_data(segment_size, Config.STEP_SIZE,
Config.NB_CLASSES, Config.DATASET_FILE_PATH,
Config.STANDARDIZATION_PATH, train_index_path,
test_index_path, validation_index_path)
# ------------------------------ Weights path ---------------------------- #
train_index_filename = os.path.basename(train_index_path).split('.')[0]
weights_to_store_foldername = train_index_filename + '_' \
+ 'batch' + str(Config.BATCH_SIZE) \
+ 'wait' + str(Config.WAIT_COUNT) \
+ 'order' + str(Config.LAYERS_ORDER_LIST[0]) \
+ 'extra' + str(Config.EXTRA_FRAMES)
fold_weights_path = os.path.join(Config.ALL_FOLDS_WEIGHTS_PATH,
weights_to_store_foldername)
if not os.path.exists(fold_weights_path):
if Config.USE_PRETRAINED_WEIGHTS == False:
os.makedirs(fold_weights_path)
elif Config.USE_PRETRAINED_WEIGHTS == True:
print('Pre-trained weights do not exist in :' +
fold_weights_path)
exit(1)
# -------------------------- Build and Train model ----------------------- #
print('----------- Training param -------------')
print(' batch_size>' + str(Config.BATCH_SIZE) + ' nb_classes>' +
str(Config.NB_CLASSES) + ' nb_epoch>' + str(Config.NB_EPOCH) +
' mclnn_layers>' + str(Config.MCLNN_LAYER_COUNT) +
' dense_layers>' + str(Config.DENSE_LAYER_COUNT) + ' norder>' +
str(Config.LAYERS_ORDER_LIST) + ' extra_frames>' +
str(Config.EXTRA_FRAMES) + ' segment_size>' +
str(segment_size + 1)
+ # plus 1 is for middle frame, considered in segmentation stage
' initial_fold>' +
str(Config.INITIAL_FOLD_ID +
1) + # plus 1 beacuse folds are zero indexed
' wait_count>' + str(Config.WAIT_COUNT) + ' split_count>' +
str(Config.SPLIT_COUNT))
if Config.USE_PRETRAINED_WEIGHTS == False:
model = mclnn_trainer.build_model(
segment_size=data_loader.train_segments.shape[1],
feature_count=data_loader.train_segments.shape[2],
pretrained_weights_path=None)
mclnn_trainer.train_model(model, data_loader, fold_weights_path)
# ------------------ Load trained weights in a new model ------------------ #
# load paths of all weights generated during training
weight_list = glob.glob(os.path.join(fold_weights_path, "*.hdf5"))
if len(weight_list) == 0:
print('Weight path is not found = ' + fold_weights_path)
return
weight_list.sort(key=os.path.getmtime)
if len(weight_list) > 1:
startup_weights = weight_list[-(Config.WAIT_COUNT + 2)]
elif len(weight_list) == 1:
startup_weights = weight_list[0]
print('----------- Weights Loaded ---------------:')
print(startup_weights)
model = mclnn_trainer.build_model(
segment_size=data_loader.train_segments.shape[1],
feature_count=data_loader.train_segments.shape[2],
pretrained_weights_path=startup_weights)
# ------------------------ Visualize a test sample --------------------- #
if is_visualization_called_flag == False and Config.SAVE_TEST_SEGMENT_PREDICTION_IMAGE == True:
visualizer = Visualizer(Config)
visualizer.visualize_weights_and_sample_test_clip(
model=model, data_loader=data_loader)
is_visualization_called_flag = True
# --------------------------- Evaluate model ------------------------------ #
fold_majority_cm, fold_probability_cm, \
fold_majority_vote_label, fold_probability_vote_label, \
fold_target_label = mclnn_trainer.evaluate_model(segment_size=segment_size,
model=model,
data_loader=data_loader)
all_folds_majority_vote_cm += fold_majority_cm
all_folds_majority_vote_label = np.append(
all_folds_majority_vote_label, fold_majority_vote_label)
all_folds_probability_vote_cm += fold_probability_cm
all_folds_probability_vote_label = np.append(
all_folds_probability_vote_label, fold_probability_vote_label)
all_folds_target_label = np.append(all_folds_target_label,
fold_target_label)
gc.collect()
print('-------------- Cross validation performance --------------')
print(Config.CLASS_NAMES)
print(all_folds_majority_vote_cm)
print(
str(Config.CROSS_VALIDATION_FOLDS_COUNT) +
'-Fold Clip-level majority-vote Accuracy ' + str(
accuracy_score(all_folds_target_label,
all_folds_majority_vote_label)))
print(Config.CLASS_NAMES)
print(all_folds_probability_vote_cm)
print(
str(Config.CROSS_VALIDATION_FOLDS_COUNT) +
'-Fold Clip-level probability-vote Accuracy ' + str(
accuracy_score(all_folds_target_label,
all_folds_probability_vote_label)))
scoref1 = f1score(all_folds_target_label,
all_folds_probability_vote_label,
average='micro')
print('F1 Score micro ' + str(scoref1))
scoref1 = f1score(all_folds_target_label,
all_folds_probability_vote_label,
average='weighted')
print('F1 Score weighted ' + str(scoref1))
# 'error': error,
# 'traceback' : traceback.format_exc()}
return Response(json.dumps(response), mimetype='application/json')
if __name__ == '__main__':
# Configuracion del registro de mensajes de seguimiento
logging.basicConfig(filename='log/chatbot.log')
logger = logging.getLogger('ai_app')
logger.setLevel(logging.DEBUG)
formatter = logging.Formatter(
'%(asctime)s - %(name)s - %(levelname)s: %(message)s',
'%d-%m-%Y %I:%M:%S %p')
console_handler = logging.StreamHandler()
ner = SubRed_NER('SUBRED_pos', 'normalization_models.pkl')
answer_manager = AnswerManager(state_machine=StateMachine(
'tree_dialog_config.json', 'trained_intentions.pkl', ner),
templates_dir='templates/',
time_per_word=0)
answer_manager.return_repeated = False
visualizer = Visualizer(ner)
with app.app_context():
db.drop_all()
db.create_all()
socketio.run(app, host=sys.argv[1], port=sys.argv[2], debug=False)
# app.run(host = '192.168.222.63', port = 8004)
try:
with open("/etc/machine-id", "r") as fh:
app.config['SECRET_KEY'] = fh.read()
except:
app.config['SECRET_KEY'] = 'secret!'
sockets = Sockets(app)
# Build pipeline
grabber = PanoramaGrabber() # config read from ~/.robovision/grabber.conf
image_recognizer = ImageRecognizer(grabber)
gameplay = Gameplay(image_recognizer)
rf = RemoteRF(gameplay, "/dev/ttyACM0")
visualizer = Visualizer(image_recognizer, framedrop=1)
recorder = Recorder(grabber)
def generator():
visualizer.enable()
queue = visualizer.get_queue()
while True:
try:
buf, resized, frame, r = queue.get_nowait()
except Empty:
sleep(0.001) # Fix this stupid thingie
continue
else:
yield b'--frame\r\nContent-Type: image/jpeg\r\n\r\n'
yield buf
yield b'\r\n\r\n'
l = Log(file_path)
# variants_count = case_statistics.get_variant_statistics(l.log)
# variants_count = \
# sorted(variants_count,
# key=lambda x: x['count'],
# reverse=True)
# print('')
# print(variants_count)
# print('')
l.filter_variants(Log.INTENSE_FILTERING)
# dfg_discovery.apply(l.log)
dfg = DFG(l.log)
Visualizer.dfg_visualizer(dfg.dfg, l.log)
# variants_count = case_statistics.get_variant_statistics(l.log)
# variants_count = \
# sorted(variants_count,
# key=lambda x: x['count'],
# reverse=True)
# print('')
# print(variants_count)
# print('')
def __init__(self):
self.graph = defaultdict(list) # default_factory() of type list
self.v = Visualizer()
self.result = ""
class Music(Default):
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.pattern_name = "Music"
CONFIGS['n_pixels'] = self.strip_length
# init visualizer
self.vis = Visualizer(CONFIGS)
# dict used to set the visualizer effect
self.effect_dict = dict(
spectrum=self.vis.visualize_spectrum,
energy=self.vis.visualize_energy,
scroll=self.vis.visualize_scroll,
)
# name of the effect to be used
self.effect = 'spectrum'
self.modifiers = dict(visualizer=self.effect, )
# attributes for the mic
self.p = None
self.stream = None
self.frames_per_buffer = int(CONFIGS['mic_rate'] / CONFIGS['fps'])
self.setup()
def setup(self):
"""
Setup stream
"""
self.p = pyaudio.PyAudio()
self.stream = self.p.open(format=pyaudio.paInt16,
channels=1,
rate=CONFIGS['mic_rate'],
input=True,
frames_per_buffer=self.frames_per_buffer)
@property
def effect(self):
return self._effect
@effect.setter
def effect(self, value):
"""
Set the effect to a certain value and change the visualization effect in the vis calss
"""
try:
ef = self.effect_dict[value]
self.vis.visualization_effect = ef
self._effect = value
except KeyError as e:
print(f"Error for key {value}\n{e}")
@property
def rate(self):
"""
Rate should always be zero here9
"""
return 0
@rate.setter
def rate(self, value):
pass
def read_audio(self):
"""
Read audio and return it
"""
try:
y = np.fromstring(self.stream.read(self.frames_per_buffer,
exception_on_overflow=False),
dtype=np.int16)
y = y.astype(np.float32)
self.stream.read(self.stream.get_read_available(),
exception_on_overflow=False)
return y
except IOError:
print('Audio buffer has overflowed')
def fill(self):
"""
Read from audio stream and set pixels
"""
# read audio input, can also be none when the mic has not started yet
output = self.read_audio()
try:
# use visualization
pixels, _ = self.vis.audio_to_rgb(output)
# Truncate values and cast to integer
pixels = np.clip(pixels, 0, 255).astype(int)
# Optional gamma correction
pixels = _gamma[pixels]
r, g, b = pixels
for idx in range(len(r)):
self.pixels[idx]['color'] = (r[idx], g[idx], b[idx], 255)
except TypeError:
pass
def stop(self):
"""
Call super method and close audio stream
"""
super(Music, self).stop()
self.stream.stop_stream()
self.stream.close()
self.p.terminate()
providers += [SSLVisionDataProvider()]
if not NO_REFBOX:
providers += [RefboxDataProvider()]
if not NO_RADIO:
providers += [Comms(HOME_TEAM)]
if CONTROL_BOTH_TEAMS:
providers += [Comms(AWAY_TEAM, True)]
providers += [Strategy(HOME_TEAM, HOME_STRATEGY)]
if CONTROL_BOTH_TEAMS:
providers += [Strategy(AWAY_TEAM, AWAY_STRATEGY)]
providers += [Visualizer()]
# Pass the providers to the coordinator
c = Coordinator(providers)
# Setup the exit handler
def stop_it(signum, frame):
c.stop_game()
signal.signal(signal.SIGINT, stop_it)
# Start the game
c.start_game()
# Exit once game is over
sys.exit()
:param iters: tweets in each partition
:return: None
'''
res = list(iters)
if not len(res): return
Visualizer.upload(res)
if __name__ == '__main__':
sc = SparkContext('local[2]', 'TwitterApp')
ssc = StreamingContext(sc, 4)
ssc.checkpoint(r'./checkpoint')
streams = ssc.socketTextStream(TCP_IP, TCP_PORT)
# windowed stream and combine data in each rdd then devide tweet correctly
windowed_stream = streams.window(4, 4).repartition(1)
corrected_stream = windowed_stream.mapPartitions(lambda iters: seperate_tweets(iters), preservesPartitioning=False)
# fileter out tweets which have no coordinates
corrected_stream = corrected_stream.filter(lambda x: re.search('\\(.+\\)', x))
sentiments = corrected_stream.filter(lambda x: len(x) > 0).\
mapPartitions(lambda iters: analyze(iters), preservesPartitioning=False)
reformat_stream = sentiments.map(lambda tup: format_data(tup))
reformat_stream.pprint()
reformat_stream.foreachRDD(lambda rdd: rdd.foreachPartition(upload_to_kibana))
Visualizer.init()
ssc.start()
ssc.awaitTermination()
Analyst.my_nlp.close()
pred_gp_mean_trajs[j] = mean_traj
pred_gp_variance_trajs[j] = variance_traj
rollout_gp_trajs[j] = state_traj
return pred_gp_mean, pred_gp_variance, rollout_gp, pred_gp_mean_trajs, pred_gp_variance_trajs, rollout_gp_trajs
if __name__ == '__main__':
import matplotlib.pyplot as plt
plt.style.use('ggplot')
from cartpole_sim import CartpoleSim
from policy import SwingUpAndBalancePolicy, RandomPolicy
from visualization import Visualizer
import cv2
vis = Visualizer(cartpole_length=1.5,
x_lim=(0.0, DELTA_T * NUM_DATAPOINTS_PER_EPOCH))
swingup_policy = SwingUpAndBalancePolicy('policy.npz')
random_policy = RandomPolicy(seed=12831)
sim = CartpoleSim(dt=DELTA_T)
# Initial training data used to train GP for the first epoch
init_state = np.array([0.01, 0.01, np.pi * 0.5, 0.1]) * rng.randn(4)
ts, state_traj, action_traj = sim_rollout(sim, random_policy,
NUM_DATAPOINTS_PER_EPOCH,
DELTA_T, init_state)
delta_state_traj = state_traj[1:] - state_traj[:-1]
train_x, train_y = make_training_data(state_traj[:-1], action_traj,
delta_state_traj)
# print(variants_count)
# print('')
# l.filter_variants(Log.INTENSE_FILTERING)
# dfg_discovery.apply(l.log)
parameters = {Parameters.NOISE_THRESHOLD:True}
im = InductiveMiner(l.log, parameters)
for place in im.net.places:
print("\nPLACE: "+place.name)
for arc in place.in_arcs:
print(arc.source.name, arc.source.label)
Visualizer.petrinet_visualizer(im.net,
im.initial_marking,
im.final_marking)
# variants_count = case_statistics.get_variant_statistics(l.log)
# variants_count = \
# sorted(variants_count,
# key=lambda x: x['count'],
# reverse=True)
# print('')
# print(variants_count)
# print('')
class Environment:
#
# Each environment should have a config and a Population
# the environment should also do the evaluation of the genomes
#
# each genome is given in a list with an ID and the actual Genome
# we will create an agent for each ID/Genome and use that to simulate
#
# each evaluation will do improve the network for X generations
# we then simulate the each network and observe its actions,
# each action will change the properties of the Agent class it was given,
# in the simulation we observe the action and what score it got from it,
# this is called an evaluation
#
# each new evaluation will be called an episode
# in each episode we store the overall score/fitness
# the agents will be visualized for Y seconds before doing another
# evaluation of the genomes.
#
# the genome fitness in our case is for an example how much food it has eaten,
# the fitness will drop extremely when the agent eats poison
#
# the fitness in the beginning of the experiment would probably be the ratio
# between food and poison
# fitness = food / poison, or fitness = food / (poison / 2)
#
def __init__(self):
self.visualizer = Visualizer()
self.render = False
self.food = []
self.poison = []
self.generation = 1
self.bounds = WORLD_BOUNDS # tuple e.g. (x, y)
self.agent = Agent()
#self.pool = None if NUM_CORES < 2 else multiprocessing.Pool(NUM_CORES)
self.init()
def get_scaled_inputs(self, agent):
pos = agent.pos
grid = []
debug_grid_pos = []
grid_radius = int(
AGENT_GRID_SIZE //
2) # get's the number of grid spaces on each side of the agent
x0grid = int(pos[0] - ((SINGLE_GRID_SIZE * grid_radius) +
(SINGLE_GRID_SIZE / 2)))
y0grid = int(pos[1] - ((SINGLE_GRID_SIZE * grid_radius) +
(SINGLE_GRID_SIZE / 2)))
#x1grid = int(x0grid + (AGENT_SIZE * AGENT_GRID_SIZE))
#y1grid = int(y0grid + (AGENT_SIZE * AGENT_GRID_SIZE))
gridsize = (SINGLE_GRID_SIZE * AGENT_GRID_SIZE)
for y in range(AGENT_GRID_SIZE):
for x in range(AGENT_GRID_SIZE):
debug_grid_pos.append((x0grid + x * SINGLE_GRID_SIZE,
y0grid + y * SINGLE_GRID_SIZE))
grid.append(0)
for idx in range(len(self.food)):
xrpos = int(self.food[idx][0] - x0grid)
yrpos = int(self.food[idx][1] - y0grid)
if xrpos >= 0 and yrpos >= 0 and xrpos < gridsize and yrpos < gridsize:
# A thing is in the grid
xrpos = int(xrpos / SINGLE_GRID_SIZE)
yrpos = int(yrpos / SINGLE_GRID_SIZE)
index = xrpos + yrpos * AGENT_GRID_SIZE
grid[index] = 1
#print(index)
for idx in range(len(self.poison)):
xrpos = int(self.poison[idx][0] - x0grid)
yrpos = int(self.poison[idx][1] - y0grid)
if xrpos >= 0 and yrpos >= 0 and xrpos < gridsize and yrpos < gridsize:
# A thing is in the grid
xrpos = int(xrpos / SINGLE_GRID_SIZE)
yrpos = int(yrpos / SINGLE_GRID_SIZE)
index = xrpos + yrpos * AGENT_GRID_SIZE
grid[index] = -1
#print(index)
if RENDER_DEBUG and self.render:
self.visualizer.drawDebug(debug_grid_pos, grid)
grid.append(agent.pos[0] / WORLD_BOUNDS[0])
grid.append(agent.pos[1] / WORLD_BOUNDS[1])
return [int(i) for i in grid]
def get_distance(self, pos1, pos2):
delta_x = pos1[0] - pos2[0]
delta_y = pos1[1] - pos2[1]
return (math.sqrt(delta_x * delta_x + delta_y * delta_y))
def step(self, actions, agent):
reward = 0
pos_x = agent.pos[0]
pos_y = agent.pos[1]
if actions[0]:
pos_x += AGENT_SPEED * actions[0]
if actions[1]:
pos_x -= AGENT_SPEED * actions[1]
if actions[2]:
pos_y += AGENT_SPEED * actions[2]
if actions[3]:
pos_y -= AGENT_SPEED * actions[3]
if LOOP_BOUNDS:
if pos_x > WORLD_BOUNDS[0]:
pos_x = 0
if pos_x < 0:
pos_x = WORLD_BOUNDS[0]
if pos_y > WORLD_BOUNDS[1]:
pos_y = 0
if pos_y < 0:
pos_y = WORLD_BOUNDS[1]
else:
if pos_x > WORLD_BOUNDS[0]:
pos_x = WORLD_BOUNDS[0]
if pos_x < 0:
pos_x = 0
if pos_y > WORLD_BOUNDS[1]:
pos_y = WORLD_BOUNDS[1]
if pos_y < 0:
pos_y = 0
agent.set_pos((int(pos_x), int(pos_y)))
toRemove_food = []
for pos in self.food:
if self.get_distance(agent.pos,
pos) < (FOOD_SIZE + AGENT_SIZE) / 2:
reward += FOOD_REWARD
toRemove_food.append(pos)
dead = False
toRemove_poison = []
for pos in self.poison:
if self.get_distance(agent.pos,
pos) < (POISON_SIZE + AGENT_SIZE) / 2:
dead = True
reward += POISON_REWARD
toRemove_poison.append(pos)
for elem in toRemove_food:
self.food.remove(elem)
for elem in toRemove_poison:
self.poison.remove(elem)
return (dead, reward)
def simulate(self, genome, net):
self.invalidate_agents = True
if self.generation % CHECKPOINT_INTERVAL == 0 and False:
self.visualizer.start_recording(
("Video/" + str(self.generation) + ".mp4"))
scores = []
score = 0
steps = 0
self.place_foods()
self.place_poisons()
self.agent.set_pos(self.get_random_pos(agent_spawn=True))
while steps < SIMULATION_TICKS:
steps += 1
old_pos = self.agent.pos
if not self.agent.alive:
break
if self.generation % CHECKPOINT_INTERVAL == 0:
self.visualizer.clear_view()
inputs = self.get_scaled_inputs(
self.agent) # a.get_scaled_inputs(self)
output = net.activate(inputs)
dead, reward = self.step(output, self.agent)
dist_moved = self.get_distance(old_pos, self.agent.pos)
if dead:
self.agent.alive = False
score += reward
if self.generation % CHECKPOINT_INTERVAL == 0 and False:
self.visualizer.update_view(self)
#self.respawn_items()
if self.generation % CHECKPOINT_INTERVAL == 0 and False:
self.visualizer.flush()
return score
def simulate_individual(self, networks):
self.invalidate_agents = True
if self.generation % CHECKPOINT_INTERVAL == 0:
self.visualizer.start_recording(
("Video/" + str(self.generation) + ".mp4"))
scores = []
rewards = {}
steps = 0
self.place_foods()
self.place_poisons()
while steps < SIMULATION_TICKS:
steps += 1
if self.generation % CHECKPOINT_INTERVAL == 0:
self.visualizer.clear_view()
for gid, genome, net in networks:
if self.invalidate_agents:
self.init_agents(networks)
self.invalidate_agents = False
if gid not in self.agents:
self.agents[gid] = Agent(gid)
a = self.agents[gid]
if not a.alive:
continue
inputs = self.get_scaled_inputs(a) # a.get_scaled_inputs(self)
output = net.activate(inputs)
dead, reward = self.step(output, a)
if dead:
a.alive = False
continue
genome.fitness += reward
if gid not in self.rewards:
self.rewards[gid] = []
self.rewards[gid].append(reward)
if self.generation % CHECKPOINT_INTERVAL == 0:
self.visualizer.update_view(self)
#self.respawn_items()
if self.generation % CHECKPOINT_INTERVAL == 0:
self.visualizer.flush()
for k in self.rewards:
scores.append(sum(self.rewards[k]))
print("Score range [{:.3f}, {:.3f}]".format(min(scores), max(scores)))
def simulate_best(self, network):
self.visualizer.start_recording(
("Video/" + str(self.generation) + " - best" + ".mp4"))
self.render = True
scores = []
rewards = {}
steps = 0
self.place_foods()
self.place_poisons()
self.agent.set_pos(self.get_random_pos(agent_spawn=True))
while steps < SIMULATION_TICKS:
steps += 1
self.visualizer.clear_view()
inputs = self.get_scaled_inputs(
self.agent) # a.get_scaled_inputs(self)
output = network.activate(inputs)
dead, reward = self.step(output, self.agent)
self.visualizer.update_view(self)
self.visualizer.flush()
def respawn_items(self):
food_count = len(self.food)
respawn_chance = (NUM_FOOD - food_count) / NUM_FOOD
if random.random() < respawn_chance and food_count < NUM_FOOD:
self.food.append(self.get_random_pos())
poison_count = len(self.poison)
respawn_chance = (NUM_POISON - poison_count) / NUM_POISON
if random.random() < respawn_chance and poison_count < NUM_POISON:
self.poison.append(self.get_random_pos())
def init(self):
self.place_foods()
self.place_poisons()
def get_random_pos(self, agent_spawn=False):
x_bound = WORLD_BOUNDS[0]
y_bound = WORLD_BOUNDS[1]
if agent_spawn:
return int(x_bound / 2), int(y_bound / 2)
while True:
rand_x = random.randrange(0, x_bound)
rand_y = random.randrange(0, y_bound)
if self.get_distance((self.bounds[0] / 2, self.bounds[1] / 2),
(rand_x, rand_y)) < SPAWN_AREA_SIZE:
continue
for f in self.food:
if self.get_distance((rand_x, rand_y), f) < MIN_SPAWN_DIST:
continue
for p in self.poison:
if self.get_distance((rand_x, rand_y), p) < MIN_SPAWN_DIST:
continue
return rand_x, rand_y
def place_foods(self):
self.food = []
for _ in range(NUM_FOOD):
self.food.append(self.get_random_pos())
def place_poisons(self):
self.poison = []
for _ in range(NUM_POISON):
self.poison.append(self.get_random_pos())
def main():
vehicle = Vehicle()
simulator = Simulator(controller=vehicle, initial_state=np.array([*POSITION0, *VELOCITY0, *ORIENTATION0, *OMEGA0]), sim_time=SIM_TIME)
simulator.simulate()
visualization = Visualizer(simulator=simulator)
print(simulator.state)
def test_zebra_contours(video_path, save_path):
visor = Visor(debuger)
visor.initializer()
navigator = BlindNavigator()
visualizer = Visualizer()
cmap = {
'LIGHT_WAIT': '等灯',
'START_FORWARD':'开始前进',
'CROSS_FORWARD':'继续前行',
'CROSS_WAIT':'有障碍物',
'ARRIVAL': '到达',
}
for iiii, data in enumerate(read_video(video_path)):
if data is None:
break
import cv2
print(data.shape)
# import pdb
# pdb.set_trace()
p_light, detected_obstacles, traffic_lights, road_mask = navigator.executor(data)
data = visor.inliner(data, save_path)
# traffic_lights = []
for j in navigator.zebra_contours:
# import pdb
# pdb.set_trace()
bx, by, bw, bh = cv2.boundingRect(j)
[bx, by, bw, bh] = visor.rescale_box([bx, by, bw, bh])
cv2.rectangle(data, (bx, by), (bx + bw, by + bh), (220, 252, 255), -1)
plt.figure()
ax = plt.gca()
ax.imshow(road_mask.astype(np.uint8))
# plt.show()
plt.savefig('test.png')
depth = navigator.depth_estimator.predict(data)
data = visualizer.plot(data, detected_obstacles, depth, navigator.is_stable, navigator.zebra_contours)
data = cv2.cvtColor(data, cv2.COLOR_BGRA2BGR)
text = cmap[navigator.state]
from PIL import Image, ImageDraw, ImageFont
cv2_im = cv2.cvtColor(data, cv2.COLOR_BGR2RGB) # cv2和PIL中颜色的hex码的储存顺序不同
pil_im = Image.fromarray(cv2_im)
draw = ImageDraw.Draw(pil_im) # 括号中为需要打印的canvas，这里就是在图片上直接打印
if 1 or p_light is not None:
# bbox, state = p_lights.get_bbox(), p_lights.get_state()
# traffic_lights.append((state, bbox))
for bbox, state, d_or_t in zip(traffic_lights.get(), traffic_lights.get_field('states'), traffic_lights.get_field('d_or_t')):
bbox = visor.rescale_box(bbox)
# assert id_ < len(navigator.traffic_light_pool.trackers), 'id_ not valid'
for i in range(thickness):
draw.rectangle([tuple(bbox[:2] - i), tuple(bbox[2:4] + i)])
label_size_ = draw.textsize('%s, %s' % (state, d_or_t), efont)
if bbox[1] - label_size_[1] >= 0:
text_origin = np.array([bbox[0], bbox[1] - label_size_[1]])
else:
text_origin = np.array([bbox[0], bbox[1] + 1])
draw.rectangle([tuple(text_origin), tuple(label_size_ + text_origin)], fill=(0x00, 0x99, 0x11))
draw.text(tuple(text_origin), '%s, %s' % (state, d_or_t), (255, 255, 255), font=efont)
text += 'frame: %d' % iiii
label_size = draw.textsize(text, cfont)
draw.rectangle(
[(0, 0), tuple(label_size)],
fill=(0x00, 0x99, 0xFF))
draw.text((0, 0), text, (255, 255, 255), font=cfont) # 第一个参数为打印的坐标，第二个为打印的文本，第三个为字体颜色，第四个为字体
data = cv2.cvtColor(np.array(pil_im), cv2.COLOR_RGB2BGR)
visor.drawer(data)