Python util.info() Examples

def list(self, url, filter=None):
        util.info("Examining url " + url)

        list_result = None
        if FILTER_URL_PARAM in url:
            list_result = self.list_movies_by_dubbing(url)
        elif not filter and DUBBING_URL_PARAM in url:
            list_result = self.list_dubbing(url)
        elif J_MOVIES_A_TO_Z_TYPE in url or J_MOVIES_GENRE in url:
            list_result = self.load_json_list(url)
        elif J_SERIES in url:
            list_result = self.list_episodes(url)
        elif J_TV_SHOWS in url or J_TV_SHOWS_MOST_POPULAR in url:
            list_result = self.list_series_letter(url)
        elif J_TV_SHOWS_RECENTLY_ADDED in url:
            list_result = self.list_recentlyadded_episodes(url)
        elif J_TV_SHOWS_A_TO_Z_TYPE in url:
            list_result = self.a_to_z(J_TV_SHOWS)
        else:
            order_by = None
            if J_MOVIES_RECENTLY_ADDED in url:
                order_by = self.order_recently_by
            list_result = self.list_videos(url, filter, order_by)
        return list_result 

def train(self, state, legal_moves, winner=False):
        assert self.memory is not None, "can't train without setting memory first"
        self.train_count += 1
        model = self.model
        if self.prev_state is None:
            # on first move, no training to do yet
            self.prev_state = state
        else:
            # add new info to replay memory
            reward = 0
            if winner == self.color:
                reward = WIN_REWARD
            elif winner == opponent[self.color]:
                reward = LOSE_REWARD
            elif winner is not False:
                raise ValueError

            self.memory.remember(self.prev_state, self.prev_move,
                                 reward, state, legal_moves, winner)

            # get an experience from memory and train on it
            if self.train_count % BATCH_SIZE == 0 or winner is not False:
                states, targets = self.memory.get_replay(
                    model, BATCH_SIZE, ALPHA)
                model.train_on_batch(states, targets) 

def get_model(self, filename=None):
        """Given a filename, load that model file; otherwise, generate a new model."""
        model = None
        if filename:
            info('attempting to load model {}'.format(filename))
            try:
                model = model_from_json(open(filename).read())
            except FileNotFoundError:
                print('could not load file {}'.format(filename))
                quit()
            print('loaded model file {}'.format(filename))
        else:
            print('no model file loaded, generating new model.')
            size = self.reversi.size ** 2
            model = Sequential()
            model.add(Dense(HIDDEN_SIZE, activation='relu', input_dim=size))
            # model.add(Dense(HIDDEN_SIZE, activation='relu'))
            model.add(Dense(size))

        model.compile(loss='mse', optimizer=optimizer)
        return model 

def tick (self, flow_now = None):
        now = time.time()
        if ((self.max_ticks is not None and self.ticks == self.max_ticks) or
            (self.max_time is not None and now > self.start + self.max_time)):
            raise Done()
        self.ticks += 1
        if flow_now is not None:
            self.flow_now = flow_now
        if ((self.tick_report is not None and
             self.ticks - self.last_report_ticks >= self.tick_report) or
            (self.flow_report is not None and self.flow_now is not None and
             ((self.flow_now - self.last_report_flow).total_seconds()
              >= self.flow_report)) or
            (self.time_report is not None and
             now - self.last_report_time >= self.time_report)):
            self.logger.info("%s",self.report())
            self.last_report_time = now
            self.last_report_ticks = self.ticks
            self.last_report_flow = self.flow_now 

def compile_file_list(self, data_dir, split, load_pose=False):
    """Creates a list of input files."""
    logging.info('data_dir: %s', data_dir)
    with gfile.Open(os.path.join(data_dir, '%s.txt' % split), 'r') as f:
      frames = f.readlines()
    subfolders = [x.split(' ')[0] for x in frames]
    frame_ids = [x.split(' ')[1][:-1] for x in frames]
    image_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '.jpg')
        for i in range(len(frames))
    ]
    cam_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '_cam.txt')
        for i in range(len(frames))
    ]
    file_lists = {}
    file_lists['image_file_list'] = image_file_list
    file_lists['cam_file_list'] = cam_file_list
    if load_pose:
      pose_file_list = [
          os.path.join(data_dir, subfolders[i], frame_ids[i] + '_pose.txt')
          for i in range(len(frames))
      ]
      file_lists['pose_file_list'] = pose_file_list
    self.steps_per_epoch = len(image_file_list) // self.batch_size
    return file_lists 

def build_inference_for_training(self):
    """Invokes depth and ego-motion networks and computes clouds if needed."""
    (self.image_stack, self.intrinsic_mat, self.intrinsic_mat_inv) = (
        self.reader.read_data())
    with tf.name_scope('egomotion_prediction'):
      self.egomotion, _ = nets.egomotion_net(self.image_stack, is_training=True,
                                             legacy_mode=self.legacy_mode)
    with tf.variable_scope('depth_prediction'):
      # Organized by ...[i][scale].  Note that the order is flipped in
      # variables in build_loss() below.
      self.disp = {}
      self.depth = {}
      if self.icp_weight > 0:
        self.cloud = {}
      for i in range(self.seq_length):
        image = self.image_stack[:, :, :, 3 * i:3 * (i + 1)]
        multiscale_disps_i, _ = nets.disp_net(image, is_training=True)
        multiscale_depths_i = [1.0 / d for d in multiscale_disps_i]
        self.disp[i] = multiscale_disps_i
        self.depth[i] = multiscale_depths_i
        if self.icp_weight > 0:
          multiscale_clouds_i = [
              project.get_cloud(d,
                                self.intrinsic_mat_inv[:, s, :, :],
                                name='cloud%d_%d' % (s, i))
              for (s, d) in enumerate(multiscale_depths_i)
          ]
          self.cloud[i] = multiscale_clouds_i
        # Reuse the same depth graph for all images.
        tf.get_variable_scope().reuse_variables()
    logging.info('disp: %s', util.info(self.disp)) 

def compile_file_list(self, data_dir, split, load_pose=False):
    """Creates a list of input files."""
    logging.info('data_dir: %s', data_dir)
    with gfile.Open(os.path.join(data_dir, '%s.txt' % split), 'r') as f:
      frames = f.readlines()
      frames = [k.rstrip() for k in frames]
    subfolders = [x.split(' ')[0] for x in frames]
    frame_ids = [x.split(' ')[1] for x in frames]
    image_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '.' +
                     self.file_extension)
        for i in range(len(frames))
    ]
    segment_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '-fseg.' +
                     self.file_extension)
        for i in range(len(frames))
    ]
    cam_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '_cam.txt')
        for i in range(len(frames))
    ]
    file_lists = {}
    file_lists['image_file_list'] = image_file_list
    file_lists['segment_file_list'] = segment_file_list
    file_lists['cam_file_list'] = cam_file_list
    if load_pose:
      pose_file_list = [
          os.path.join(data_dir, subfolders[i], frame_ids[i] + '_pose.txt')
          for i in range(len(frames))
      ]
      file_lists['pose_file_list'] = pose_file_list
    self.steps_per_epoch = len(image_file_list) // self.batch_size
    return file_lists 

def compile_file_list(self, data_dir, split, load_pose=False):
    """Creates a list of input files."""
    logging.info('data_dir: %s', data_dir)
    with gfile.Open(os.path.join(data_dir, '%s.txt' % split), 'r') as f:
      frames = f.readlines()
    subfolders = [x.split(' ')[0] for x in frames]
    frame_ids = [x.split(' ')[1][:-1] for x in frames]
    image_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '.jpg')
        for i in range(len(frames))
    ]
    cam_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '_cam.txt')
        for i in range(len(frames))
    ]
    file_lists = {}
    file_lists['image_file_list'] = image_file_list
    file_lists['cam_file_list'] = cam_file_list
    if load_pose:
      pose_file_list = [
          os.path.join(data_dir, subfolders[i], frame_ids[i] + '_pose.txt')
          for i in range(len(frames))
      ]
      file_lists['pose_file_list'] = pose_file_list
    self.steps_per_epoch = len(image_file_list) // self.batch_size
    return file_lists 

def build_inference_for_training(self):
    """Invokes depth and ego-motion networks and computes clouds if needed."""
    (self.image_stack, self.intrinsic_mat, self.intrinsic_mat_inv) = (
        self.reader.read_data())
    with tf.name_scope('egomotion_prediction'):
      self.egomotion, _ = nets.egomotion_net(self.image_stack, is_training=True,
                                             legacy_mode=self.legacy_mode)
    with tf.variable_scope('depth_prediction'):
      # Organized by ...[i][scale].  Note that the order is flipped in
      # variables in build_loss() below.
      self.disp = {}
      self.depth = {}
      if self.icp_weight > 0:
        self.cloud = {}
      for i in range(self.seq_length):
        image = self.image_stack[:, :, :, 3 * i:3 * (i + 1)]
        multiscale_disps_i, _ = nets.disp_net(image, is_training=True)
        multiscale_depths_i = [1.0 / d for d in multiscale_disps_i]
        self.disp[i] = multiscale_disps_i
        self.depth[i] = multiscale_depths_i
        if self.icp_weight > 0:
          multiscale_clouds_i = [
              project.get_cloud(d,
                                self.intrinsic_mat_inv[:, s, :, :],
                                name='cloud%d_%d' % (s, i))
              for (s, d) in enumerate(multiscale_depths_i)
          ]
          self.cloud[i] = multiscale_clouds_i
        # Reuse the same depth graph for all images.
        tf.get_variable_scope().reuse_variables()
    logging.info('disp: %s', util.info(self.disp)) 

def compile_file_list(self, data_dir, split, load_pose=False):
    """Creates a list of input files."""
    logging.info('data_dir: %s', data_dir)
    with gfile.Open(os.path.join(data_dir, '%s.txt' % split), 'r') as f:
      frames = f.readlines()
      frames = [k.rstrip() for k in frames]
    subfolders = [x.split(' ')[0] for x in frames]
    frame_ids = [x.split(' ')[1] for x in frames]
    image_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '.' +
                     self.file_extension)
        for i in range(len(frames))
    ]
    segment_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '-fseg.' +
                     self.file_extension)
        for i in range(len(frames))
    ]
    cam_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '_cam.txt')
        for i in range(len(frames))
    ]
    file_lists = {}
    file_lists['image_file_list'] = image_file_list
    file_lists['segment_file_list'] = segment_file_list
    file_lists['cam_file_list'] = cam_file_list
    if load_pose:
      pose_file_list = [
          os.path.join(data_dir, subfolders[i], frame_ids[i] + '_pose.txt')
          for i in range(len(frames))
      ]
      file_lists['pose_file_list'] = pose_file_list
    self.steps_per_epoch = len(image_file_list) // self.batch_size
    return file_lists 

def compile_file_list(self, data_dir, split, load_pose=False):
    """Creates a list of input files."""
    logging.info('data_dir: %s', data_dir)
    with gfile.Open(os.path.join(data_dir, '%s.txt' % split), 'r') as f:
      frames = f.readlines()
    subfolders = [x.split(' ')[0] for x in frames]
    frame_ids = [x.split(' ')[1][:-1] for x in frames]
    image_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '.jpg')
        for i in range(len(frames))
    ]
    cam_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '_cam.txt')
        for i in range(len(frames))
    ]
    file_lists = {}
    file_lists['image_file_list'] = image_file_list
    file_lists['cam_file_list'] = cam_file_list
    if load_pose:
      pose_file_list = [
          os.path.join(data_dir, subfolders[i], frame_ids[i] + '_pose.txt')
          for i in range(len(frames))
      ]
      file_lists['pose_file_list'] = pose_file_list
    self.steps_per_epoch = len(image_file_list) // self.batch_size
    return file_lists 

def build_inference_for_training(self):
    """Invokes depth and ego-motion networks and computes clouds if needed."""
    (self.image_stack, self.intrinsic_mat, self.intrinsic_mat_inv) = (
        self.reader.read_data())
    with tf.name_scope('egomotion_prediction'):
      self.egomotion, _ = nets.egomotion_net(self.image_stack, is_training=True,
                                             legacy_mode=self.legacy_mode)
    with tf.variable_scope('depth_prediction'):
      # Organized by ...[i][scale].  Note that the order is flipped in
      # variables in build_loss() below.
      self.disp = {}
      self.depth = {}
      if self.icp_weight > 0:
        self.cloud = {}
      for i in range(self.seq_length):
        image = self.image_stack[:, :, :, 3 * i:3 * (i + 1)]
        multiscale_disps_i, _ = nets.disp_net(image, is_training=True)
        multiscale_depths_i = [1.0 / d for d in multiscale_disps_i]
        self.disp[i] = multiscale_disps_i
        self.depth[i] = multiscale_depths_i
        if self.icp_weight > 0:
          multiscale_clouds_i = [
              project.get_cloud(d,
                                self.intrinsic_mat_inv[:, s, :, :],
                                name='cloud%d_%d' % (s, i))
              for (s, d) in enumerate(multiscale_depths_i)
          ]
          self.cloud[i] = multiscale_clouds_i
        # Reuse the same depth graph for all images.
        tf.get_variable_scope().reuse_variables()
    logging.info('disp: %s', util.info(self.disp)) 

def compile_file_list(self, data_dir, split, load_pose=False):
    """Creates a list of input files."""
    logging.info('data_dir: %s', data_dir)
    with gfile.Open(os.path.join(data_dir, '%s.txt' % split), 'r') as f:
      frames = f.readlines()
      frames = [k.rstrip() for k in frames]
    subfolders = [x.split(' ')[0] for x in frames]
    frame_ids = [x.split(' ')[1] for x in frames]
    image_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '.' +
                     self.file_extension)
        for i in range(len(frames))
    ]
    segment_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '-fseg.' +
                     self.file_extension)
        for i in range(len(frames))
    ]
    cam_file_list = [
        os.path.join(data_dir, subfolders[i], frame_ids[i] + '_cam.txt')
        for i in range(len(frames))
    ]
    file_lists = {}
    file_lists['image_file_list'] = image_file_list
    file_lists['segment_file_list'] = segment_file_list
    file_lists['cam_file_list'] = cam_file_list
    if load_pose:
      pose_file_list = [
          os.path.join(data_dir, subfolders[i], frame_ids[i] + '_pose.txt')
          for i in range(len(frames))
      ]
      file_lists['pose_file_list'] = pose_file_list
    self.steps_per_epoch = len(image_file_list) // self.batch_size
    return file_lists 

def timing (func, logger = None):
    start = time.time()
    ret = func()
    util.info("Ran %s in %s" % (func, elapsed(start)),logger=logger)
    return ret 

def policy(self, state, legal_moves):
        """The policy of picking an action based on their weights."""
        if not legal_moves:
            return None

        if not self.minimax_enabled:
            # don't use minimax if we're in learning mode
            best_move, _ = best_move_val(
                self.model.predict(numpify(state)),
                legal_moves
            )
            return best_move
        else:
            next_states = {self.reversi.next_state(
                state, move): move for move in legal_moves}
            move_scores = []
            for s in next_states.keys():
                score = self.minimax(s)
                move_scores.append((score, s))
                info('{}: {}'.format(next_states[s], score))

            best_val = -float('inf')
            best_move = None
            for each in move_scores:
                if each[0] > best_val:
                    best_val = each[0]
                    best_move = next_states[each[1]]

            assert best_move is not None
            return best_move 

def set_epsilon(self, val):
        self.epsilon = val
        if not self.learning_enabled:
            info('Warning -- set_epsilon() was called when learning was not enabled.') 

def evalSchedules(self):
        if not self.scanRunning() and not self.isPlaying():
            notified = False
            util.info("SOSAC Loading subscriptions")
            subs = self.get_subs()
            new_items = False
            for url, sub in subs.iteritems():
                if xbmc.abortRequested:
                    util.info("SOSAC Exiting")
                    return
                if sub['type'] == sosac.LIBRARY_TYPE_TVSHOW:
                    if self.scanRunning() or self.isPlaying():
                        self.cache.delete("subscription.last_run")
                        return
                    refresh = int(sub['refresh'])
                    if refresh > 0:
                        next_check = sub['last_run'] + (refresh * 3600 * 24)
                        if next_check < time.time():
                            if not notified:
                                self.showNotification(
                                    'Subscription', 'Chcecking')
                                notified = True
                            util.debug("SOSAC Refreshing " + url)
                            new_items |= self.run_custom({
                                'action': sosac.LIBRARY_ACTION_ADD,
                                'type': sosac.LIBRARY_TYPE_TVSHOW,
                                'update': True,
                                'url': url,
                                'name': sub['name'],
                                'refresh': sub['refresh']
                            })
                            self.sleep(3000)
                        else:
                            n = (next_check - time.time()) / 3600
                            util.debug("SOSAC Skipping " + url +
                                       " , next check in %dh" % n)
            if new_items:
                xbmc.executebuiltin('UpdateLibrary(video)')
            notified = False
        else:
            util.info("SOSAC Scan skipped") 

def service(self):
        util.info("SOSAC Service Started")
        try:
            sleep_time = int(self.getSetting("start_sleep_time")) * 1000 * 60 * 60
        except:
            sleep_time = self.sleep_time
            pass

        self.sleep(sleep_time)

        try:
            self.last_run = float(self.cache.get("subscription.last_run"))
        except:
            self.last_run = time.time()
            self.cache.set("subscription.last_run", str(self.last_run))
            pass

        if not xbmc.abortRequested and time.time() > self.last_run:
            self.evalSchedules()

        while not xbmc.abortRequested:
            # evaluate subsciptions every 10 minutes
            if(time.time() > self.last_run + 600):
                self.evalSchedules()
                self.last_run = time.time()
                self.cache.set("subscription.last_run", str(self.last_run))
            self.sleep(self.sleep_time)
        util.info("SOSAC Shutdown") 

def request_last_update(self, url):
        util.debug('request: %s' % url)
        lastmod = None
        req = urllib2.Request(url)
        req.add_header('User-Agent', util.UA)
        try:
            response = urllib2.urlopen(req)
            lastmod = datetime.datetime(*response.info().getdate('Last-Modified')[:6]).strftime(
                '%d.%m.%Y')
            response.close()
        except urllib2.HTTPError, error:
            util.debug(error.read())
            error.close() 

def __init__(self,
               data_dir=None,
               is_training=True,
               learning_rate=0.0002,
               beta1=0.9,
               reconstr_weight=0.85,
               smooth_weight=0.05,
               ssim_weight=0.15,
               icp_weight=0.0,
               batch_size=4,
               img_height=128,
               img_width=416,
               seq_length=3,
               legacy_mode=False):
    self.data_dir = data_dir
    self.is_training = is_training
    self.learning_rate = learning_rate
    self.reconstr_weight = reconstr_weight
    self.smooth_weight = smooth_weight
    self.ssim_weight = ssim_weight
    self.icp_weight = icp_weight
    self.beta1 = beta1
    self.batch_size = batch_size
    self.img_height = img_height
    self.img_width = img_width
    self.seq_length = seq_length
    self.legacy_mode = legacy_mode

    logging.info('data_dir: %s', data_dir)
    logging.info('learning_rate: %s', learning_rate)
    logging.info('beta1: %s', beta1)
    logging.info('smooth_weight: %s', smooth_weight)
    logging.info('ssim_weight: %s', ssim_weight)
    logging.info('icp_weight: %s', icp_weight)
    logging.info('batch_size: %s', batch_size)
    logging.info('img_height: %s', img_height)
    logging.info('img_width: %s', img_width)
    logging.info('seq_length: %s', seq_length)
    logging.info('legacy_mode: %s', legacy_mode)

    if self.is_training:
      self.reader = reader.DataReader(self.data_dir, self.batch_size,
                                      self.img_height, self.img_width,
                                      self.seq_length, NUM_SCALES)
      self.build_train_graph()
    else:
      self.build_depth_test_graph()
      self.build_egomotion_test_graph()

    # At this point, the model is ready.  Print some info on model params.
    util.count_parameters() 

def read_data(self):
    """Provides images and camera intrinsics."""
    with tf.name_scope('data_loading'):
      with tf.name_scope('enqueue_paths'):
        seed = random.randint(0, 2**31 - 1)
        self.file_lists = self.compile_file_list(self.data_dir, 'train')
        image_paths_queue = tf.train.string_input_producer(
            self.file_lists['image_file_list'], seed=seed, shuffle=True)
        cam_paths_queue = tf.train.string_input_producer(
            self.file_lists['cam_file_list'], seed=seed, shuffle=True)
        img_reader = tf.WholeFileReader()
        _, image_contents = img_reader.read(image_paths_queue)
        image_seq = tf.image.decode_jpeg(image_contents)

      with tf.name_scope('load_intrinsics'):
        cam_reader = tf.TextLineReader()
        _, raw_cam_contents = cam_reader.read(cam_paths_queue)
        rec_def = []
        for _ in range(9):
          rec_def.append([1.0])
        raw_cam_vec = tf.decode_csv(raw_cam_contents, record_defaults=rec_def)
        raw_cam_vec = tf.stack(raw_cam_vec)
        intrinsics = tf.reshape(raw_cam_vec, [3, 3])

      with tf.name_scope('convert_image'):
        image_seq = self.preprocess_image(image_seq)  # Converts to float.

      with tf.name_scope('image_augmentation'):
        image_seq = self.augment_image_colorspace(image_seq)

      image_stack = self.unpack_images(image_seq)

      with tf.name_scope('image_augmentation_scale_crop'):
        image_stack, intrinsics = self.augment_images_scale_crop(
            image_stack, intrinsics, self.img_height, self.img_width)

      with tf.name_scope('multi_scale_intrinsics'):
        intrinsic_mat = self.get_multi_scale_intrinsics(intrinsics,
                                                        self.num_scales)
        intrinsic_mat.set_shape([self.num_scales, 3, 3])
        intrinsic_mat_inv = tf.matrix_inverse(intrinsic_mat)
        intrinsic_mat_inv.set_shape([self.num_scales, 3, 3])

      with tf.name_scope('batching'):
        image_stack, intrinsic_mat, intrinsic_mat_inv = (
            tf.train.shuffle_batch(
                [image_stack, intrinsic_mat, intrinsic_mat_inv],
                batch_size=self.batch_size,
                capacity=QUEUE_SIZE + QUEUE_BUFFER * self.batch_size,
                min_after_dequeue=QUEUE_SIZE))
        logging.info('image_stack: %s', util.info(image_stack))
    return image_stack, intrinsic_mat, intrinsic_mat_inv 

def __init__(self,
               data_dir=None,
               is_training=True,
               learning_rate=0.0002,
               beta1=0.9,
               reconstr_weight=0.85,
               smooth_weight=0.05,
               ssim_weight=0.15,
               icp_weight=0.0,
               batch_size=4,
               img_height=128,
               img_width=416,
               seq_length=3,
               legacy_mode=False):
    self.data_dir = data_dir
    self.is_training = is_training
    self.learning_rate = learning_rate
    self.reconstr_weight = reconstr_weight
    self.smooth_weight = smooth_weight
    self.ssim_weight = ssim_weight
    self.icp_weight = icp_weight
    self.beta1 = beta1
    self.batch_size = batch_size
    self.img_height = img_height
    self.img_width = img_width
    self.seq_length = seq_length
    self.legacy_mode = legacy_mode

    logging.info('data_dir: %s', data_dir)
    logging.info('learning_rate: %s', learning_rate)
    logging.info('beta1: %s', beta1)
    logging.info('smooth_weight: %s', smooth_weight)
    logging.info('ssim_weight: %s', ssim_weight)
    logging.info('icp_weight: %s', icp_weight)
    logging.info('batch_size: %s', batch_size)
    logging.info('img_height: %s', img_height)
    logging.info('img_width: %s', img_width)
    logging.info('seq_length: %s', seq_length)
    logging.info('legacy_mode: %s', legacy_mode)

    if self.is_training:
      self.reader = reader.DataReader(self.data_dir, self.batch_size,
                                      self.img_height, self.img_width,
                                      self.seq_length, NUM_SCALES)
      self.build_train_graph()
    else:
      self.build_depth_test_graph()
      self.build_egomotion_test_graph()

    # At this point, the model is ready.  Print some info on model params.
    util.count_parameters() 

def read_data(self):
    """Provides images and camera intrinsics."""
    with tf.name_scope('data_loading'):
      with tf.name_scope('enqueue_paths'):
        seed = random.randint(0, 2**31 - 1)
        self.file_lists = self.compile_file_list(self.data_dir, 'train')
        image_paths_queue = tf.train.string_input_producer(
            self.file_lists['image_file_list'], seed=seed, shuffle=True)
        cam_paths_queue = tf.train.string_input_producer(
            self.file_lists['cam_file_list'], seed=seed, shuffle=True)
        img_reader = tf.WholeFileReader()
        _, image_contents = img_reader.read(image_paths_queue)
        image_seq = tf.image.decode_jpeg(image_contents)

      with tf.name_scope('load_intrinsics'):
        cam_reader = tf.TextLineReader()
        _, raw_cam_contents = cam_reader.read(cam_paths_queue)
        rec_def = []
        for _ in range(9):
          rec_def.append([1.0])
        raw_cam_vec = tf.decode_csv(raw_cam_contents, record_defaults=rec_def)
        raw_cam_vec = tf.stack(raw_cam_vec)
        intrinsics = tf.reshape(raw_cam_vec, [3, 3])

      with tf.name_scope('convert_image'):
        image_seq = self.preprocess_image(image_seq)  # Converts to float.

      with tf.name_scope('image_augmentation'):
        image_seq = self.augment_image_colorspace(image_seq)

      image_stack = self.unpack_images(image_seq)

      with tf.name_scope('image_augmentation_scale_crop'):
        image_stack, intrinsics = self.augment_images_scale_crop(
            image_stack, intrinsics, self.img_height, self.img_width)

      with tf.name_scope('multi_scale_intrinsics'):
        intrinsic_mat = self.get_multi_scale_intrinsics(intrinsics,
                                                        self.num_scales)
        intrinsic_mat.set_shape([self.num_scales, 3, 3])
        intrinsic_mat_inv = tf.matrix_inverse(intrinsic_mat)
        intrinsic_mat_inv.set_shape([self.num_scales, 3, 3])

      with tf.name_scope('batching'):
        image_stack, intrinsic_mat, intrinsic_mat_inv = (
            tf.train.shuffle_batch(
                [image_stack, intrinsic_mat, intrinsic_mat_inv],
                batch_size=self.batch_size,
                capacity=QUEUE_SIZE + QUEUE_BUFFER * self.batch_size,
                min_after_dequeue=QUEUE_SIZE))
        logging.info('image_stack: %s', util.info(image_stack))
    return image_stack, intrinsic_mat, intrinsic_mat_inv 

def __init__(self,
               data_dir=None,
               is_training=True,
               learning_rate=0.0002,
               beta1=0.9,
               reconstr_weight=0.85,
               smooth_weight=0.05,
               ssim_weight=0.15,
               icp_weight=0.0,
               batch_size=4,
               img_height=128,
               img_width=416,
               seq_length=3,
               legacy_mode=False):
    self.data_dir = data_dir
    self.is_training = is_training
    self.learning_rate = learning_rate
    self.reconstr_weight = reconstr_weight
    self.smooth_weight = smooth_weight
    self.ssim_weight = ssim_weight
    self.icp_weight = icp_weight
    self.beta1 = beta1
    self.batch_size = batch_size
    self.img_height = img_height
    self.img_width = img_width
    self.seq_length = seq_length
    self.legacy_mode = legacy_mode

    logging.info('data_dir: %s', data_dir)
    logging.info('learning_rate: %s', learning_rate)
    logging.info('beta1: %s', beta1)
    logging.info('smooth_weight: %s', smooth_weight)
    logging.info('ssim_weight: %s', ssim_weight)
    logging.info('icp_weight: %s', icp_weight)
    logging.info('batch_size: %s', batch_size)
    logging.info('img_height: %s', img_height)
    logging.info('img_width: %s', img_width)
    logging.info('seq_length: %s', seq_length)
    logging.info('legacy_mode: %s', legacy_mode)

    if self.is_training:
      self.reader = reader.DataReader(self.data_dir, self.batch_size,
                                      self.img_height, self.img_width,
                                      self.seq_length, NUM_SCALES)
      self.build_train_graph()
    else:
      self.build_depth_test_graph()
      self.build_egomotion_test_graph()

    # At this point, the model is ready.  Print some info on model params.
    util.count_parameters() 

def read_data(self):
    """Provides images and camera intrinsics."""
    with tf.name_scope('data_loading'):
      with tf.name_scope('enqueue_paths'):
        seed = random.randint(0, 2**31 - 1)
        self.file_lists = self.compile_file_list(self.data_dir, 'train')
        image_paths_queue = tf.train.string_input_producer(
            self.file_lists['image_file_list'], seed=seed, shuffle=True)
        cam_paths_queue = tf.train.string_input_producer(
            self.file_lists['cam_file_list'], seed=seed, shuffle=True)
        img_reader = tf.WholeFileReader()
        _, image_contents = img_reader.read(image_paths_queue)
        image_seq = tf.image.decode_jpeg(image_contents)

      with tf.name_scope('load_intrinsics'):
        cam_reader = tf.TextLineReader()
        _, raw_cam_contents = cam_reader.read(cam_paths_queue)
        rec_def = []
        for _ in range(9):
          rec_def.append([1.0])
        raw_cam_vec = tf.decode_csv(raw_cam_contents, record_defaults=rec_def)
        raw_cam_vec = tf.stack(raw_cam_vec)
        intrinsics = tf.reshape(raw_cam_vec, [3, 3])

      with tf.name_scope('convert_image'):
        image_seq = self.preprocess_image(image_seq)  # Converts to float.

      with tf.name_scope('image_augmentation'):
        image_seq = self.augment_image_colorspace(image_seq)

      image_stack = self.unpack_images(image_seq)

      with tf.name_scope('image_augmentation_scale_crop'):
        image_stack, intrinsics = self.augment_images_scale_crop(
            image_stack, intrinsics, self.img_height, self.img_width)

      with tf.name_scope('multi_scale_intrinsics'):
        intrinsic_mat = self.get_multi_scale_intrinsics(intrinsics,
                                                        self.num_scales)
        intrinsic_mat.set_shape([self.num_scales, 3, 3])
        intrinsic_mat_inv = tf.matrix_inverse(intrinsic_mat)
        intrinsic_mat_inv.set_shape([self.num_scales, 3, 3])

      with tf.name_scope('batching'):
        image_stack, intrinsic_mat, intrinsic_mat_inv = (
            tf.train.shuffle_batch(
                [image_stack, intrinsic_mat, intrinsic_mat_inv],
                batch_size=self.batch_size,
                capacity=QUEUE_SIZE + QUEUE_BUFFER * self.batch_size,
                min_after_dequeue=QUEUE_SIZE))
        logging.info('image_stack: %s', util.info(image_stack))
    return image_stack, intrinsic_mat, intrinsic_mat_inv 

def main():
    logger.info("Checking your OS version...")
    if util.check_os():
        logger.info("OK")
    else:
        logger.critical("You must use Ubuntu 14.04")

    if util.not_sudo():
        logger.critical("Restart script as root")

    logger.info("Installing packages...")
    if util.install_packages():
        logger.info("OK")
    else:
        logger.critical("Failed to install packages")

    logger.info("Applying sysctl parameters...")
    if util.setup_sysctl():
        logger.info("OK")
    else:
        logger.critical("Failed to apply sysctl parameters")

    logger.info("Creating random passwords...")
    if util.setup_passwords():
        logger.info("OK")
    else:
        logger.critical("Failed to create random passwords")

    logger.info("Other config files...")
    if util.cp_configs():
        logger.info("OK")
    else:
        logger.critical("Fail")

    logger.info("Adding script to rc.local...")
    if util.setup_vpn():
        logger.info("OK")
    else:
        logger.critical("Failed adding script to rc.local")

    logger.info("Installing web UI...")
    if util.webui():
        logger.info("OK")
    else:
        logger.critical("Failed installing web UI")

    util.info()