def __init__(self, checkpoint_dir=os.path.join(os.path.dirname(__file__), '..', LOG_DIR, 'checkpoints'),
gpu_id=GPU_ID, profiler: Profiler = None):
os.environ['CUDA_VISIBLE_DEVICES'] = str(gpu_id)
tf.Graph().as_default()
self.imagePlaceholder = tf.placeholder(tf.uint8, shape=(None, CROP_SIZE, CROP_SIZE, 3))
self.prevLstmState = tuple([tf.placeholder(tf.float32, shape=(None, LSTM_SIZE)) for _ in range(4)])
self.batch_size = tf.placeholder(tf.int32, shape=())
self.outputs, self.state1, self.state2 = network.inference(
self.imagePlaceholder, num_unrolls=1, batch_size=self.batch_size, train=False,
prevLstmState=self.prevLstmState)
self.sess = tf_util.Session()
self.sess.run(tf.global_variables_initializer())
ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
if ckpt is None:
raise IOError(
('Checkpoint model could not be found. '
'Did you download the pretrained weights? '
'Download them here: http://bit.ly/2L5deYF and read the Model section of the Readme.'))
tf_util.restore(self.sess, ckpt.model_checkpoint_path)
self.tracked_data = {}
self.time = 0
self.total_forward_count = -1
self._profiler: Profiler = profiler_pipe(profiler)
self._re3_crop_profiler = "re3 cropping image"
self._re3_crop2_profiler = "re3 cropping image 2"
self._re3_sess_profiler = "re3 session run"
self._re3_sess2_profiler = "re3 session run 2"
def __init__(self, gpu_id=GPU_ID):
os.environ['CUDA_VISIBLE_DEVICES'] = str(gpu_id)
basedir = os.path.dirname(__file__)
tf.Graph().as_default()
self.imagePlaceholder = tf.placeholder(tf.uint8,
shape=(None, CROP_SIZE,
CROP_SIZE, 3))
self.prevLstmState = tuple([
tf.placeholder(tf.float32, shape=(None, LSTM_SIZE))
for _ in range(4)
])
self.batch_size = tf.placeholder(tf.int32, shape=())
self.outputs, self.state1, self.state2 = network.inference(
self.imagePlaceholder,
num_unrolls=1,
batch_size=self.batch_size,
train=False,
prevLstmState=self.prevLstmState)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
self.sess = tf.Session(config=config)
ckpt = tf.train.get_checkpoint_state(
os.path.join(basedir, '..', LOG_DIR, 'checkpoints'))
if ckpt is None:
raise IOError((
'Checkpoint model could not be found. '
'Did you download the pretrained weights? '
'Download them here: https://goo.gl/NWGXGM and read the Model section of the Readme.'
))
tf_util.restore(self.sess, ckpt.model_checkpoint_path)
self.tracked_data = {}
self.time = 0
self.total_forward_count = -1
def create_tracker(self):
tracker = Re3Tracker(self.sess, tracked_data=self.tracked_data, lock=self.lock, reuse=self.is_initialized)
if not self.is_initialized:
basedir = os.path.dirname(__file__)
ckpt = tf.train.get_checkpoint_state(os.path.join(basedir, '..', LOG_DIR, 'checkpoints'))
tf_util.restore(self.sess, ckpt.model_checkpoint_path)
self.is_initialized = True
return tracker
def create_tracker(self):
tracker = Re3Tracker(self.sess,
tracked_data=self.tracked_data,
lock=self.lock,
reuse=self.is_initialized)
if not self.is_initialized:
ckpt = tf.train.get_checkpoint_state(self.checkpoint_dir)
tf_util.restore(self.sess, ckpt.model_checkpoint_path)
self.is_initialized = True
return tracker
def __init__(self, gpu_id=GPU_ID):
os.environ['CUDA_VISIBLE_DEVICES'] = str(gpu_id)
if getattr(sys, 'frozen', False):
# application_path = os.path.dirname(sys.executable)
basedir = os.path.dirname(sys.executable)
elif __file__:
# application_path = os.path.dirname(__file__)
basedir = os.path.dirname(__file__)
tf.Graph().as_default()
self.imagePlaceholder = tf.placeholder(tf.uint8,
shape=(None, CROP_SIZE,
CROP_SIZE, 3))
self.prevLstmState = tuple([
tf.placeholder(tf.float32, shape=(None, LSTM_SIZE))
for _ in range(4)
])
self.batch_size = tf.placeholder(tf.int32, shape=())
self.outputs, self.state1, self.state2, self.conv_layers1 = network.inference(
self.imagePlaceholder,
num_unrolls=1,
batch_size=self.batch_size,
train=False,
prevLstmState=self.prevLstmState)
self.conv_layers = network.returnConvLayers(
self.imagePlaceholder,
num_unrolls=1,
batch_size=self.batch_size,
train=False,
prevLstmState=self.prevLstmState)
self.sess = tf_util.Session()
self.sess.run(tf.global_variables_initializer())
ckpt = tf.train.get_checkpoint_state(
os.path.join(basedir, '..', LOG_DIR, 'checkpoints'))
if ckpt is None:
raise IOError((
'Checkpoint model could not be found. '
'Did you download the pretrained weights? '
'Download them here: http://bit.ly/2L5deYF and read the Model section of the Readme.'
))
tf_util.restore(self.sess, ckpt.model_checkpoint_path)
self.tracked_data = {}
self.time = 0
self.total_forward_count = -1
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
dataset = Dataset(sess, delta, 1, port, debug)
forwardNetworkImagePlaceholder = tf.placeholder(tf.uint8,
shape=(2, CROP_SIZE,
CROP_SIZE, 3))
prevLstmState = tuple(
[tf.placeholder(tf.float32, shape=(1, LSTM_SIZE)) for _ in range(4)])
initialLstmState = tuple([np.zeros((1, LSTM_SIZE)) for _ in range(4)])
networkOutputs, state1, state2 = network.inference(
forwardNetworkImagePlaceholder,
num_unrolls=1,
train=False,
prevLstmState=prevLstmState,
reuse=False)
dataset.initialize_tf_placeholders(forwardNetworkImagePlaceholder,
prevLstmState, networkOutputs, state1,
state2)
init = tf.global_variables_initializer()
sess.run(init)
ckpt = tf.train.get_checkpoint_state(LOG_DIR + '/checkpoints')
if ckpt and ckpt.model_checkpoint_path:
tf_util.restore(sess, ckpt.model_checkpoint_path)
startIter = int(ckpt.model_checkpoint_path.split('-')[-1])
print('Restored', startIter)
iteration = 0
while True:
iteration += 1
print('iteration', iteration)
dataset.get_data_sequence()