def main(args): network = importlib.import_module(args.model_def) subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S') log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir) if not os.path.isdir( log_dir): # Create the log directory if it doesn't exist os.makedirs(log_dir) model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir) if not os.path.isdir( model_dir): # Create the model directory if it doesn't exist os.makedirs(model_dir) # Write arguments to a text file facenet.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt')) # Store some git revision info in a text file in the log directory # src_path,_ = os.path.split(os.path.realpath(__file__)) # facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv)) np.random.seed(seed=args.seed) # Fetch dataets # VGGface2 print('Fetch VGGface2 as source dataset at {}'.format( args.vggface2_train_dir)) src_train_set = facenet.get_dataset(args.vggface2_train_dir) # validation_set = facenet.get_dataset(args.vggface2_val_dir) # COX S2V print('Fetch COX-S2V as target dataset at {}'.format(args.cox_video_dir)) cox_dataset = cox.cox_data(args.cox_still_dir, args.cox_video_dir, args.cox_pairs) train_folds = [0, 1, 2] evaluation_folds = [3, 4, 5, 6, 7, 8, 9] # cox_train_list = cox_dataset.get_dataset(train_folds) # cox_val_list = list(itertools.chain.from_iterable(fold_list[3:10])) tgt_train_set = cox_dataset.get_dataset(train_folds, video_only=True) # tgt_val_set = cox.get_video_dataset(args.cox_video_dir, cox_val_list) cox_paths, cox_issame = cox_dataset.get_pairs(evaluation_folds) print('Model directory: %s' % model_dir) print('Log directory: %s' % log_dir) if args.pretrained_model: print('Pre-trained model: %s' % os.path.expanduser(args.pretrained_model)) if args.lfw_dir: print('LFW directory: %s' % args.lfw_dir) # Read the file containing the pairs used for testing pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs)) # Get the paths for the corresponding images lfw_paths, actual_issame = lfw.get_paths( os.path.expanduser(args.lfw_dir), pairs) # Get the paths for embeddings projection # Get the paths for embeddings projection lfw_proj_paths, lfw_proj_labels = lfw.get_paths_from_file( args.lfw_dir, args.lfw_projection) cox_proj_paths, cox_proj_labels = cox_dataset.get_paths_from_file( args.cox_projection) # Combine projection paths projection_paths = lfw_proj_paths + cox_proj_paths proj_labels = lfw_proj_labels + cox_proj_labels # Create label map if does not exist metadata_filename = 'meta.tsv' emb_dir = os.path.join(os.path.expanduser(log_dir), 'emb') if not os.path.isdir( emb_dir): # Create the log directory if it doesn't exist os.makedirs(emb_dir) with open(os.path.join(emb_dir, metadata_filename), "w") as meta_file: csvWriter = csv.writer(meta_file, delimiter='\t') csvWriter.writerows(np.array([proj_labels]).T) with tf.Graph().as_default(): tf.set_random_seed(args.seed) global_step = tf.Variable(0, trainable=False) # Placeholder for the learning rate learning_rate_placeholder = tf.placeholder(tf.float32, name='learning_rate') batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size') phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train') image_paths_placeholder = tf.placeholder(tf.string, shape=(None, 4), name='image_paths') labels_placeholder = tf.placeholder(tf.int64, shape=(None, 4), name='labels') input_queue = data_flow_ops.FIFOQueue(capacity=100000, dtypes=[tf.string, tf.int64], shapes=[(4, ), (4, )], shared_name=None, name=None) enqueue_op = input_queue.enqueue_many( [image_paths_placeholder, labels_placeholder]) nrof_preprocess_threads = 4 images_and_labels = [] for _ in range(nrof_preprocess_threads): filenames, label = input_queue.dequeue() images = [] for filename in tf.unstack(filenames): file_contents = tf.read_file(filename) image = tf.image.decode_image(file_contents, channels=3) if args.random_crop: image = tf.random_crop( image, [args.image_size, args.image_size, 3]) else: image = tf.image.resize_image_with_crop_or_pad( image, args.image_size, args.image_size) if args.random_flip: image = tf.image.random_flip_left_right(image) #pylint: disable=no-member image.set_shape((args.image_size, args.image_size, 3)) images.append(tf.image.per_image_standardization(image)) images_and_labels.append([images, label]) image_batch, labels_batch = tf.train.batch_join( images_and_labels, batch_size=batch_size_placeholder, shapes=[(args.image_size, args.image_size, 3), ()], enqueue_many=True, capacity=4 * nrof_preprocess_threads * args.batch_size, allow_smaller_final_batch=True) image_batch = tf.identity(image_batch, 'image_batch') image_batch = tf.identity(image_batch, 'input') labels_batch = tf.identity(labels_batch, 'label_batch') # Build the inference graph prelogits, _ = network.inference( image_batch, args.keep_probability, phase_train=phase_train_placeholder, bottleneck_layer_size=args.embedding_size, weight_decay=args.weight_decay) embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings') # Split embeddings into anchor, positive and negative and calculate triplet loss anchor, positive, negative, target = tf.unstack( tf.reshape(embeddings, [-1, 4, args.embedding_size]), 4, 1) loss, adv_loss, triplet_loss = adversarialloss.quadruplets_loss( anchor, positive, negative, target, args.alpha, args.lamb, args.zeta) # triplet_loss = tripletloss.triplet_loss(anchor, positive, negative, args.alpha) # adv_loss = adversarialloss.adversarial_loss(anchor, target, args.alpha) learning_rate = tf.train.exponential_decay( learning_rate_placeholder, global_step, args.learning_rate_decay_epochs * args.epoch_size, args.learning_rate_decay_factor, staircase=True) tf.summary.scalar('learning_rate', learning_rate) # Calculate the total losses regularization_losses = tf.get_collection( tf.GraphKeys.REGULARIZATION_LOSSES) total_loss = tf.add_n([loss] + regularization_losses, name='total_loss') reg_loss = tf.add_n(regularization_losses, name='total_loss') # Build a Graph that trains the model with one batch of examples and updates the model parameters train_op = facenet.train(total_loss, global_step, args.optimizer, learning_rate, args.moving_average_decay, tf.global_variables()) # Create a saver saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3) # Build the summary operation based on the TF collection of Summaries. # summary_op = tf.summary.merge_all() # Start running operations on the Graph. gpu_options = tf.GPUOptions( per_process_gpu_memory_fraction=args.gpu_memory_fraction) sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) # Initialize variables sess.run(tf.global_variables_initializer(), feed_dict={phase_train_placeholder: True}) sess.run(tf.local_variables_initializer(), feed_dict={phase_train_placeholder: True}) summary_writer = tf.summary.FileWriter(log_dir, sess.graph) coord = tf.train.Coordinator() tf.train.start_queue_runners(coord=coord, sess=sess) with sess.as_default(): if args.pretrained_model: print('Restoring pretrained model: %s' % args.pretrained_model) saver.restore(sess, os.path.expanduser(args.pretrained_model)) # Training and validation loop epoch = 0 while epoch < args.max_nrof_epochs: step = sess.run(global_step, feed_dict=None) epoch = step // args.epoch_size # save_embeddings(sess, projection_paths, epoch, embeddings, labels_batch, image_paths_placeholder, # labels_placeholder, # batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, # args.batch_size, emb_dir, args.embedding_size, tag='lfw') # # # Evaluate on COX # evaluate(sess, cox_paths, embeddings, labels_batch, image_paths_placeholder, # labels_placeholder, # batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, # enqueue_op, # cox_issame, args.batch_size, # args.lfw_nrof_folds, log_dir, step, summary_writer, args.embedding_size, # tag='cox') # # if args.lfw_dir: # evaluate(sess, lfw_paths, embeddings, labels_batch, image_paths_placeholder, labels_placeholder, # batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, actual_issame, args.batch_size, # args.lfw_nrof_folds, log_dir, step, summary_writer, args.embedding_size) # Train for one epoch train(args, sess, src_train_set, tgt_train_set, epoch, image_paths_placeholder, labels_placeholder, labels_batch, batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, input_queue, global_step, embeddings, total_loss, triplet_loss, adv_loss, reg_loss, train_op, summary_writer, args.learning_rate_schedule_file, args.embedding_size) # Save variables and the metagraph if it doesn't exist already save_variables_and_metagraph(sess, saver, summary_writer, model_dir, subdir, step) # Evaluate on LFW return model_dir
def main(args): network = importlib.import_module(args.model_def) image_size = (args.image_size, args.image_size) subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S') # 用当前时间作为文件夹名称 log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir) # log保存路径 if not os.path.isdir( log_dir): # Create the log directory if it doesn't exist os.makedirs(log_dir) model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir) # 训练好的模型的保存路径 if not os.path.isdir( model_dir): # Create the model directory if it doesn't exist os.makedirs(model_dir) stat_file_name = os.path.join(log_dir, 'stat.h5') # Write arguments to a text file 这个文件保存一些训练时候用到的超参数信息 facenet.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt')) # Store some git revision info in a text file in the log directory # 保存一些版本信息 src_path, _ = os.path.split(os.path.realpath(__file__)) facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv)) np.random.seed(seed=args.seed) random.seed(args.seed) dataset = facenet.get_dataset(args.data_dir) # 训练数据路径 if args.filter_filename: dataset = filter_dataset(dataset, os.path.expanduser(args.filter_filename), args.filter_percentile, args.filter_min_nrof_images_per_class) # 数据集划分为训练集和验证集 if args.validation_set_split_ratio > 0.0: train_set, val_set = facenet.split_dataset( dataset, args.validation_set_split_ratio, args.min_nrof_val_images_per_class, 'SPLIT_IMAGES') else: train_set, val_set = dataset, [] # 分类的数量,即训练集中的文件夹数量 nrof_classes = len(train_set) print('Model directory: %s' % model_dir) print('Log directory: %s' % log_dir) pretrained_model = None # 预训练模型 if args.pretrained_model: pretrained_model = os.path.expanduser(args.pretrained_model) print('Pre-trained model: %s' % pretrained_model) # 测试用的lfw数据的路径 if args.lfw_dir: print('LFW directory: %s' % args.lfw_dir) # Read the file containing the pairs used for testing pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs)) # Get the paths for the corresponding images lfw_paths, actual_issame = lfw.get_paths( os.path.expanduser(args.lfw_dir), pairs) with tf.Graph().as_default(): tf.set_random_seed(args.seed) global_step = tf.Variable(0, trainable=False) # Get a list of image paths and their labels # 获取训练图像路径和对应的标签 image_list, label_list = facenet.get_image_paths_and_labels(train_set) assert len(image_list) > 0, 'The training set should not be empty' # 用于验证的图像的路径和对应的标签 val_image_list, val_label_list = facenet.get_image_paths_and_labels( val_set) # Create a queue that produces indices into the image_list and label_list labels = ops.convert_to_tensor(label_list, dtype=tf.int32) range_size = array_ops.shape(labels)[0] index_queue = tf.train.range_input_producer(range_size, num_epochs=None, shuffle=True, seed=None, capacity=32) index_dequeue_op = index_queue.dequeue_many( args.batch_size * args.epoch_size, 'index_dequeue') learning_rate_placeholder = tf.placeholder(tf.float32, name='learning_rate') batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size') phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train') image_paths_placeholder = tf.placeholder(tf.string, shape=(None, 1), name='image_paths') labels_placeholder = tf.placeholder(tf.int32, shape=(None, 1), name='labels') control_placeholder = tf.placeholder(tf.int32, shape=(None, 1), name='control') nrof_preprocess_threads = 4 # 先入先出队列 input_queue = data_flow_ops.FIFOQueue( capacity=2000000, dtypes=[tf.string, tf.int32, tf.int32], shapes=[(1, ), (1, ), (1, )], shared_name=None, name=None) enqueue_op = input_queue.enqueue_many( [image_paths_placeholder, labels_placeholder, control_placeholder], name='enqueue_op') image_batch, label_batch = facenet.create_input_pipeline( input_queue, image_size, nrof_preprocess_threads, batch_size_placeholder) image_batch = tf.identity(image_batch, 'image_batch') image_batch = tf.identity(image_batch, 'input') label_batch = tf.identity(label_batch, 'label_batch') print('Number of classes in training set: %d' % nrof_classes) print('Number of examples in training set: %d' % len(image_list)) print('Number of classes in validation set: %d' % len(val_set)) print('Number of examples in validation set: %d' % len(val_image_list)) print('Building training graph') # Build the inference graph prelogits, _ = network.inference( image_batch, args.keep_probability, phase_train=phase_train_placeholder, bottleneck_layer_size=args.embedding_size, weight_decay=args.weight_decay) logits = slim.fully_connected( prelogits, len(train_set), activation_fn=None, weights_initializer=slim.initializers.xavier_initializer(), weights_regularizer=slim.l2_regularizer(args.weight_decay), scope='Logits', reuse=False) # 标准化 embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings') # Norm for the prelogits eps = 1e-4 prelogits_norm = tf.reduce_mean( tf.norm(tf.abs(prelogits) + eps, ord=args.prelogits_norm_p, axis=1)) tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_norm * args.prelogits_norm_loss_factor) # Add center loss prelogits_center_loss, _ = facenet.center_loss(prelogits, label_batch, args.center_loss_alfa, nrof_classes) tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_center_loss * args.center_loss_factor) learning_rate = tf.train.exponential_decay( learning_rate_placeholder, global_step, args.learning_rate_decay_epochs * args.epoch_size, args.learning_rate_decay_factor, staircase=True) tf.summary.scalar('learning_rate', learning_rate) # Calculate the average cross entropy loss across the batch cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits( labels=label_batch, logits=logits, name='cross_entropy_per_example') cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy') tf.add_to_collection('losses', cross_entropy_mean) correct_prediction = tf.cast( tf.equal(tf.argmax(logits, 1), tf.cast(label_batch, tf.int64)), tf.float32) accuracy = tf.reduce_mean(correct_prediction) # Calculate the total losses regularization_losses = tf.get_collection( tf.GraphKeys.REGULARIZATION_LOSSES) total_loss = tf.add_n([cross_entropy_mean] + regularization_losses, name='total_loss') # Build a Graph that trains the model with one batch of examples and updates the model parameters train_op = facenet.train(total_loss, global_step, args.optimizer, learning_rate, args.moving_average_decay, tf.global_variables(), args.log_histograms) # Create a saver saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=1, save_relative_paths=True) # Build the summary operation based on the TF collection of Summaries. summary_op = tf.summary.merge_all() # Start running operations on the Graph. gpu_options = tf.GPUOptions( per_process_gpu_memory_fraction=args.gpu_memory_fraction) sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False)) sess.run(tf.global_variables_initializer()) sess.run(tf.local_variables_initializer()) summary_writer = tf.summary.FileWriter(log_dir, sess.graph) coord = tf.train.Coordinator() tf.train.start_queue_runners(coord=coord, sess=sess) with sess.as_default(): # if pretrained_model: # print('Restoring pretrained model: %s' % pretrained_model) # saver.restore(sess, tf.train.latest_checkpoint(pretrained_model)) if args.pretrained_model: print('Restoring pretrained model: %s' % args.pretrained_model) ckpt = tf.train.get_checkpoint_state(args.pretrained_model) if ckpt and ckpt.model_checkpoint_path: saver.restore(sess, ckpt.model_checkpoint_path) # Training and validation loop print('Running training') nrof_steps = args.max_nrof_epochs * args.epoch_size nrof_val_samples = int( math.ceil(args.max_nrof_epochs / args.validate_every_n_epochs) ) # Validate every validate_every_n_epochs as well as in the last epoch stat = { 'loss': np.zeros((nrof_steps, ), np.float32), 'center_loss': np.zeros((nrof_steps, ), np.float32), 'reg_loss': np.zeros((nrof_steps, ), np.float32), 'xent_loss': np.zeros((nrof_steps, ), np.float32), 'prelogits_norm': np.zeros((nrof_steps, ), np.float32), 'accuracy': np.zeros((nrof_steps, ), np.float32), 'val_loss': np.zeros((nrof_val_samples, ), np.float32), 'val_xent_loss': np.zeros((nrof_val_samples, ), np.float32), 'val_accuracy': np.zeros((nrof_val_samples, ), np.float32), 'lfw_accuracy': np.zeros((args.max_nrof_epochs, ), np.float32), 'lfw_valrate': np.zeros((args.max_nrof_epochs, ), np.float32), 'learning_rate': np.zeros((args.max_nrof_epochs, ), np.float32), 'time_train': np.zeros((args.max_nrof_epochs, ), np.float32), 'time_validate': np.zeros((args.max_nrof_epochs, ), np.float32), 'time_evaluate': np.zeros((args.max_nrof_epochs, ), np.float32), 'prelogits_hist': np.zeros((args.max_nrof_epochs, 1000), np.float32), } for epoch in range(1, args.max_nrof_epochs + 1): step = sess.run(global_step, feed_dict=None) # Train for one epoch t = time.time() cont = train( args, sess, epoch, image_list, label_list, index_dequeue_op, enqueue_op, image_paths_placeholder, labels_placeholder, learning_rate_placeholder, phase_train_placeholder, batch_size_placeholder, control_placeholder, global_step, total_loss, train_op, summary_op, summary_writer, regularization_losses, args.learning_rate_schedule_file, stat, cross_entropy_mean, accuracy, learning_rate, prelogits, prelogits_center_loss, args.random_rotate, args.random_crop, args.random_flip, prelogits_norm, args.prelogits_hist_max, args.use_fixed_image_standardization) stat['time_train'][epoch - 1] = time.time() - t if not cont: break t = time.time() if len(val_image_list) > 0 and ( (epoch - 1) % args.validate_every_n_epochs == args.validate_every_n_epochs - 1 or epoch == args.max_nrof_epochs): validate(args, sess, epoch, val_image_list, val_label_list, enqueue_op, image_paths_placeholder, labels_placeholder, control_placeholder, phase_train_placeholder, batch_size_placeholder, stat, total_loss, regularization_losses, cross_entropy_mean, accuracy, args.validate_every_n_epochs, args.use_fixed_image_standardization) stat['time_validate'][epoch - 1] = time.time() - t # Save variables and the metagraph if it doesn't exist already save_variables_and_metagraph(sess, saver, summary_writer, model_dir, subdir, epoch) # Evaluate on LFW t = time.time() if args.lfw_dir: evaluate(sess, enqueue_op, image_paths_placeholder, labels_placeholder, phase_train_placeholder, batch_size_placeholder, control_placeholder, embeddings, label_batch, lfw_paths, actual_issame, args.lfw_batch_size, args.lfw_nrof_folds, log_dir, step, summary_writer, stat, epoch, args.lfw_distance_metric, args.lfw_subtract_mean, args.lfw_use_flipped_images, args.use_fixed_image_standardization) stat['time_evaluate'][epoch - 1] = time.time() - t print('Saving statistics') with h5py.File(stat_file_name, 'w') as f: for key, value in stat.items(): f.create_dataset(key, data=value) return model_dir
def main(args): np.random.seed(seed=args.seed) network = importlib.import_module(args.model_def) chokepoint_dataset = chokepoint.chokepoint_data(args.still_dir, args.video_dir, args.chokepoint_pairs) fold_list = [([0, 1], [2, 3, 4]), ([1, 2], [3, 4, 0]), ([2, 3], [4, 0, 1]), ([3, 4], [0, 1, 2]), ([4, 0], [1, 2, 3])] for fold_idx in range(5): print('Fold: {}'.format(fold_idx)) train_folds, evaluation_folds = fold_list[fold_idx] # Train set chokepoint_train_set = chokepoint_dataset.get_dataset(train_folds) chokepoint1_paths, chokepoint1_issame = chokepoint_dataset.get_pairs( train_folds) # del chokepoint1_paths[-4:] # del chokepoint1_issame[-2:] # Validation set chokepoint2_paths, chokepoint2_issame = chokepoint_dataset.get_pairs( evaluation_folds) subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S') log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), 'fold{}_'.format(fold_idx) + subdir) if not os.path.isdir( log_dir): # Create the log directory if it doesn't exist os.makedirs(log_dir) model_dir = os.path.join(os.path.expanduser(args.models_base_dir), 'fold{}_'.format(fold_idx) + subdir) if not os.path.isdir( model_dir): # Create the model directory if it doesn't exist os.makedirs(model_dir) # Write arguments to a text file facenet.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt')) # Store some git revision info in a text file in the log directory src_path, _ = os.path.split(os.path.realpath(__file__)) facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv)) print('Model directory: %s' % model_dir) print('Log directory: %s' % log_dir) if args.pretrained_model: print('Pre-trained model: %s' % os.path.expanduser(args.pretrained_model)) if args.lfw_dir: print('LFW directory: %s' % args.lfw_dir) # Read the file containing the pairs used for testing pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs)) # Get the paths for the corresponding images lfw_paths, lfw_issame = lfw.get_paths( os.path.expanduser(args.lfw_dir), pairs) with tf.Graph().as_default(): tf.set_random_seed(args.seed) global_step = tf.Variable(0, trainable=False) # Placeholder for the learning rate learning_rate_placeholder = tf.placeholder(tf.float32, name='learning_rate') batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size') phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train') image_paths_placeholder = tf.placeholder(tf.string, shape=(None, 3), name='image_paths') labels_placeholder = tf.placeholder(tf.int64, shape=(None, 3), name='labels') input_queue = data_flow_ops.FIFOQueue(capacity=100000, dtypes=[tf.string, tf.int64], shapes=[(3, ), (3, )], shared_name=None, name=None) enqueue_op = input_queue.enqueue_many( [image_paths_placeholder, labels_placeholder]) nrof_preprocess_threads = 4 images_and_labels = [] for _ in range(nrof_preprocess_threads): filenames, label = input_queue.dequeue() images = [] for filename in tf.unstack(filenames): file_contents = tf.read_file(filename) image = tf.image.decode_image(file_contents, channels=3) if args.random_crop: image = tf.random_crop( image, [args.image_size, args.image_size, 3]) else: image = tf.image.resize_image_with_crop_or_pad( image, args.image_size, args.image_size) if args.random_flip: image = tf.image.random_flip_left_right(image) #pylint: disable=no-member image.set_shape((args.image_size, args.image_size, 3)) images.append(tf.image.per_image_standardization(image)) images_and_labels.append([images, label]) image_batch, labels_batch = tf.train.batch_join( images_and_labels, batch_size=batch_size_placeholder, shapes=[(args.image_size, args.image_size, 3), ()], enqueue_many=True, capacity=4 * nrof_preprocess_threads * args.batch_size, allow_smaller_final_batch=True) image_batch = tf.identity(image_batch, 'image_batch') image_batch = tf.identity(image_batch, 'input') labels_batch = tf.identity(labels_batch, 'label_batch') # Build the inference graph prelogits, _ = network.inference( image_batch, args.keep_probability, phase_train=phase_train_placeholder, bottleneck_layer_size=args.embedding_size, weight_decay=args.weight_decay) embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings') # Split embeddings into anchor, positive and negative and calculate triplet loss anchor, positive, negative = tf.unstack( tf.reshape(embeddings, [-1, 3, args.embedding_size]), 3, 1) triplet_loss = facenet.triplet_loss(anchor, positive, negative, args.alpha) learning_rate = tf.train.exponential_decay( learning_rate_placeholder, global_step, args.learning_rate_decay_epochs * args.epoch_size, args.learning_rate_decay_factor, staircase=True) tf.summary.scalar('learning_rate', learning_rate) # Calculate the total losses regularization_losses = tf.get_collection( tf.GraphKeys.REGULARIZATION_LOSSES) total_loss = tf.add_n([triplet_loss] + regularization_losses, name='total_loss') # Build a Graph that trains the model with one batch of examples and updates the model parameters train_op = facenet.train(total_loss, global_step, args.optimizer, learning_rate, args.moving_average_decay, tf.global_variables()) # Create a saver saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3) # Build the summary operation based on the TF collection of Summaries. # summary_op = tf.summary.merge_all() # Start running operations on the Graph. gpu_options = tf.GPUOptions( per_process_gpu_memory_fraction=args.gpu_memory_fraction) sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) # Initialize variables sess.run(tf.global_variables_initializer(), feed_dict={phase_train_placeholder: True}) sess.run(tf.local_variables_initializer(), feed_dict={phase_train_placeholder: True}) summary_writer = tf.summary.FileWriter(log_dir, sess.graph) coord = tf.train.Coordinator() tf.train.start_queue_runners(coord=coord, sess=sess) with sess.as_default(): if args.pretrained_model: print('Restoring pretrained model: %s' % args.pretrained_model) saver.restore(sess, os.path.expanduser(args.pretrained_model)) # Training and validation loop epoch = 0 while epoch < args.max_nrof_epochs: step = sess.run(global_step, feed_dict=None) epoch = step // args.epoch_size # Evaluate on COX evaluate(sess, chokepoint1_paths, embeddings, labels_batch, image_paths_placeholder, labels_placeholder, batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, chokepoint1_issame, 100, 2, log_dir, step, summary_writer, args.embedding_size, tag='chokepoint_train') evaluate(sess, chokepoint2_paths, embeddings, labels_batch, image_paths_placeholder, labels_placeholder, batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, chokepoint2_issame, 100, 3, log_dir, step, summary_writer, args.embedding_size, tag='chokepoint_eval') # Evaluate on LFW # if args.lfw_dir: # lfw_result = evaluate(sess, lfw_paths, embeddings, labels_batch, image_paths_placeholder, labels_placeholder, # batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, # lfw_issame, args.batch_size, # args.lfw_nrof_folds, log_dir, step, summary_writer, args.embedding_size, tag='lfw') # Train for one epoch train(args, sess, chokepoint_train_set, epoch, image_paths_placeholder, labels_placeholder, labels_batch, batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, input_queue, global_step, embeddings, total_loss, train_op, summary_writer, args.learning_rate_schedule_file, args.embedding_size) # Save variables and the metagraph if it doesn't exist already save_variables_and_metagraph(sess, saver, summary_writer, model_dir, subdir, step) return model_dir
def main(args): img_mean = np.array([134.10714722, 102.52040863, 87.15436554]) img_stddev = np.sqrt( np.array([3941.30175781, 2856.94287109, 2519.35791016])) vae_def = importlib.import_module(args.vae_def) vae = vae_def.Vae(args.latent_var_size) gen_image_size = vae.get_image_size() subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S') model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir) if not os.path.isdir( model_dir): # Create the model directory if it doesn't exist os.makedirs(model_dir) log_file_name = os.path.join(model_dir, 'logs.h5') # Write arguments to a text file facenet.write_arguments_to_file(args, os.path.join(model_dir, 'arguments.txt')) # Store some git revision info in a text file in the log directory src_path, _ = os.path.split(os.path.realpath(__file__)) facenet.store_revision_info(src_path, model_dir, ' '.join(sys.argv)) with tf.Graph().as_default(): tf.set_random_seed(args.seed) global_step = tf.Variable(0, trainable=False) train_set = facenet.get_dataset(args.data_dir) image_list, _ = facenet.get_image_paths_and_labels(train_set) # Create the input queue input_queue = tf.train.string_input_producer(image_list, shuffle=True) nrof_preprocess_threads = 4 image_per_thread = [] for _ in range(nrof_preprocess_threads): file_contents = tf.read_file(input_queue.dequeue()) image = tf.image.decode_image(file_contents, channels=3) image = tf.image.resize_image_with_crop_or_pad( image, args.input_image_size, args.input_image_size) image.set_shape((args.input_image_size, args.input_image_size, 3)) image = tf.cast(image, tf.float32) #pylint: disable=no-member image_per_thread.append([image]) images = tf.train.batch_join(image_per_thread, batch_size=args.batch_size, capacity=4 * nrof_preprocess_threads * args.batch_size, allow_smaller_final_batch=False) # Normalize images_norm = (images - img_mean) / img_stddev # Resize to appropriate size for the encoder images_norm_resize = tf.image.resize_images( images_norm, (gen_image_size, gen_image_size)) # Create encoder network mean, log_variance = vae.encoder(images_norm_resize, True) epsilon = tf.random_normal((tf.shape(mean)[0], args.latent_var_size)) std = tf.exp(log_variance / 2) latent_var = mean + epsilon * std # Create decoder network reconstructed_norm = vae.decoder(latent_var, True) # Un-normalize reconstructed = (reconstructed_norm * img_stddev) + img_mean # Create reconstruction loss if args.reconstruction_loss_type == 'PLAIN': images_resize = tf.image.resize_images( images, (gen_image_size, gen_image_size)) reconstruction_loss = tf.reduce_mean( tf.reduce_sum(tf.pow(images_resize - reconstructed, 2))) elif args.reconstruction_loss_type == 'PERCEPTUAL': network = importlib.import_module(args.model_def) reconstructed_norm_resize = tf.image.resize_images( reconstructed_norm, (args.input_image_size, args.input_image_size)) # Stack images from both the input batch and the reconstructed batch in a new tensor shp = [-1] + images_norm.get_shape().as_list()[1:] input_images = tf.reshape( tf.stack([images_norm, reconstructed_norm_resize], axis=0), shp) _, end_points = network.inference(input_images, 1.0, phase_train=False, bottleneck_layer_size=128, weight_decay=0.0) # Get a list of feature names to use for loss terms feature_names = args.loss_features.replace(' ', '').split(',') # Calculate L2 loss between original and reconstructed images in feature space reconstruction_loss_list = [] for feature_name in feature_names: feature_flat = slim.flatten(end_points[feature_name]) image_feature, reconstructed_feature = tf.unstack(tf.reshape( feature_flat, [2, args.batch_size, -1]), num=2, axis=0) reconstruction_loss = tf.reduce_mean(tf.reduce_sum( tf.pow(image_feature - reconstructed_feature, 2)), name=feature_name + '_loss') reconstruction_loss_list.append(reconstruction_loss) # Sum up the losses in for the different features reconstruction_loss = tf.add_n(reconstruction_loss_list, 'reconstruction_loss') else: pass # Create KL divergence loss kl_loss = kl_divergence_loss(mean, log_variance) kl_loss_mean = tf.reduce_mean(kl_loss) total_loss = args.alfa * kl_loss_mean + args.beta * reconstruction_loss learning_rate = tf.train.exponential_decay( args.initial_learning_rate, global_step, args.learning_rate_decay_steps, args.learning_rate_decay_factor, staircase=True) # Calculate gradients and make sure not to include parameters for the perceptual loss model opt = tf.train.AdamOptimizer(learning_rate) grads = opt.compute_gradients(total_loss, var_list=get_variables_to_train()) # Apply gradients apply_gradient_op = opt.apply_gradients(grads, global_step=global_step) with tf.control_dependencies([apply_gradient_op]): train_op = tf.no_op(name='train') # Create a saver saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3) facenet_saver = tf.train.Saver(get_facenet_variables_to_restore()) # Start running operations on the Graph gpu_memory_fraction = 1.0 gpu_options = tf.GPUOptions( per_process_gpu_memory_fraction=gpu_memory_fraction) sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False)) sess.run(tf.global_variables_initializer()) sess.run(tf.local_variables_initializer()) coord = tf.train.Coordinator() tf.train.start_queue_runners(coord=coord, sess=sess) with sess.as_default(): if args.reconstruction_loss_type == 'PERCEPTUAL': if not args.pretrained_model: raise ValueError( 'A pretrained model must be specified when using perceptual loss' ) pretrained_model_exp = os.path.expanduser( args.pretrained_model) print('Restoring pretrained model: %s' % pretrained_model_exp) facenet_saver.restore(sess, pretrained_model_exp) log = { 'total_loss': np.zeros((0, ), np.float), 'reconstruction_loss': np.zeros((0, ), np.float), 'kl_loss': np.zeros((0, ), np.float), 'learning_rate': np.zeros((0, ), np.float), } step = 0 print('Running training') while step < args.max_nrof_steps: start_time = time.time() step += 1 save_state = step > 0 and (step % args.save_every_n_steps == 0 or step == args.max_nrof_steps) if save_state: _, reconstruction_loss_, kl_loss_mean_, total_loss_, learning_rate_, rec_ = sess.run( [ train_op, reconstruction_loss, kl_loss_mean, total_loss, learning_rate, reconstructed ]) img = facenet.put_images_on_grid(rec_, shape=(16, 8)) misc.imsave( os.path.join(model_dir, 'reconstructed_%06d.png' % step), img) else: _, reconstruction_loss_, kl_loss_mean_, total_loss_, learning_rate_ = sess.run( [ train_op, reconstruction_loss, kl_loss_mean, total_loss, learning_rate ]) log['total_loss'] = np.append(log['total_loss'], total_loss_) log['reconstruction_loss'] = np.append( log['reconstruction_loss'], reconstruction_loss_) log['kl_loss'] = np.append(log['kl_loss'], kl_loss_mean_) log['learning_rate'] = np.append(log['learning_rate'], learning_rate_) duration = time.time() - start_time print( 'Step: %d \tTime: %.3f \trec_loss: %.3f \tkl_loss: %.3f \ttotal_loss: %.3f' % (step, duration, reconstruction_loss_, kl_loss_mean_, total_loss_)) if save_state: print('Saving checkpoint file') checkpoint_path = os.path.join(model_dir, 'model.ckpt') saver.save(sess, checkpoint_path, global_step=step, write_meta_graph=False) print('Saving log') with h5py.File(log_file_name, 'w') as f: for key, value in iteritems(log): f.create_dataset(key, data=value)
def main(args): #此处导入的是:models.inception_resnet_v1模型,以后再看怎么更改模型 network = importlib.import_module(args.model_def) #用当前日期来命名模型 subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S') #日志保存在c:\\users\\Administrator\logs\facenet\ 文件夹里 log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir) if not os.path.isdir( log_dir): # Create the log directory if it doesn't exist os.makedirs(log_dir) #没有日志文件就创建一个 model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir) if not os.path.isdir( model_dir): # Create the model directory if it doesn't exist os.makedirs(model_dir) # 把参数写在日志文件中 facenet.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt')) # Store some git revision info in a text file in the log directory src_path, _ = os.path.split(os.path.realpath(__file__)) #arg_string:'E:/facenet/train_tripletloss.py' output_dir:'C:\\Users\\Administrator/logs/facenet\\20180314-181556' facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv)) np.random.seed(seed=args.seed) #获取数据集,train_set是包含文件路径与标签的集合 #先输入一个父路径 path:'E:/facenet/data/lfw_160',接着输入每个子路径 # 输出:一个list,每个元素是一个ImageClass,里边包含图片地址的list(image_paths)以及对应的人名(name)[以后可能会直接调用这几个属性] train_set = facenet.get_dataset(args.data_dir) print('Model directory: %s' % model_dir) print('Log directory: %s' % log_dir) if args.pretrained_model: #用在判断是否有预训练模型,但是如果有,怎么加载呢? print('Pre-trained model: %s' % os.path.expanduser(args.pretrained_model)) if args.lfw_dir: print('LFW directory: %s' % args.lfw_dir) # Read the file containing the pairs used for testing pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs)) # Get the paths for the corresponding images lfw_paths, actual_issame = lfw.get_paths( os.path.expanduser(args.lfw_dir), pairs, args.lfw_file_ext) #建立图 #with语句适用于对资源进行访问的场合,确保使用过程中是否发生异常都会执行必要嘚瑟“清理”操作 with tf.Graph().as_default(): tf.set_random_seed(args.seed) global_step = tf.Variable(0, trainable=False) # 学习率 Placeholder for the learning rate learning_rate_placeholder = tf.placeholder(tf.float32, name='learning_rate') #批大小 batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size') #用于判断是训练还是测试 phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train') #图像路径 image_paths_placeholder = tf.placeholder(tf.string, shape=(None, 3), name='image_paths') # 图像标签 labels_placeholder = tf.placeholder(tf.int64, shape=(None, 3), name='labels') #新建一个队列,数据流操作,fifo先入先出 input_queue = data_flow_ops.FIFOQueue(capacity=100000, dtypes=[tf.string, tf.int64], shapes=[(3, ), (3, )], shared_name=None, name=None) #enqueue_many返回的是一个操作 enqueue_op = input_queue.enqueue_many( [image_paths_placeholder, labels_placeholder]) nrof_preprocess_threads = 4 images_and_labels = [] for _ in range(nrof_preprocess_threads): filenames, label = input_queue.dequeue() images = [] for filename in tf.unstack(filenames): file_contents = tf.read_file(filename) image = tf.image.decode_image(file_contents, channels=3) if args.random_crop: image = tf.random_crop( image, [args.image_size, args.image_size, 3]) else: image = tf.image.resize_image_with_crop_or_pad( image, args.image_size, args.image_size) if args.random_flip: image = tf.image.random_flip_left_right(image) #pylint: disable=no-member image.set_shape((args.image_size, args.image_size, 3)) images.append(tf.image.per_image_standardization(image)) images_and_labels.append([images, label]) image_batch, labels_batch = tf.train.batch_join( images_and_labels, batch_size=batch_size_placeholder, shapes=[(args.image_size, args.image_size, 3), ()], enqueue_many=True, capacity=4 * nrof_preprocess_threads * args.batch_size, allow_smaller_final_batch=True) image_batch = tf.identity(image_batch, 'image_batch') image_batch = tf.identity(image_batch, 'input') labels_batch = tf.identity(labels_batch, 'label_batch') # Build the inference (构造计算图) #其中prelogits是最后一层的输出 prelogits, _ = network.inference( image_batch, args.keep_probability, phase_train=phase_train_placeholder, bottleneck_layer_size=args.embedding_size, weight_decay=args.weight_decay) #L2正则化(范化)函数 # embeddings = tf.nn.l2_normalize(输入向量, L2范化的维数(取0(列L2范化)或1(行L2范化)), 泛化的最小值边界, name='embeddings') embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings') # Split embeddings into anchor, positive and negative and calculate triplet loss anchor, positive, negative = tf.unstack( tf.reshape(embeddings, [-1, 3, args.embedding_size]), 3, 1) triplet_loss = facenet.triplet_loss(anchor, positive, negative, args.alpha) #将指数衰减应用在学习率上 learning_rate = tf.train.exponential_decay( learning_rate_placeholder, global_step, args.learning_rate_decay_epochs * args.epoch_size, args.learning_rate_decay_factor, staircase=True) tf.summary.scalar('learning_rate', learning_rate) # 计算损失 regularization_losses = tf.get_collection( tf.GraphKeys.REGULARIZATION_LOSSES) #构建L2正则化 total_loss = tf.add_n([triplet_loss] + regularization_losses, name='total_loss') # Build a Graph that trains the model with one batch of examples and updates the model parameters # 确定优化方法并根据损失函数求梯度,在这里,每更行一次参数,global_step会加1 train_op = facenet.train(total_loss, global_step, args.optimizer, learning_rate, args.moving_average_decay, tf.global_variables()) # Create a saver创建一个saver用来保存或者从内存中回复一个模型参数 saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3) # Build the summary operation based on the TF collection of Summaries. summary_op = tf.summary.merge_all() # Start running operations on the Graph.能够在GPU上分配的最大内存 gpu_options = tf.GPUOptions( per_process_gpu_memory_fraction=args.gpu_memory_fraction) sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) # Initialize variables sess.run(tf.global_variables_initializer(), feed_dict={phase_train_placeholder: True}) sess.run(tf.local_variables_initializer(), feed_dict={phase_train_placeholder: True}) #写log文件 summary_writer = tf.summary.FileWriter(log_dir, sess.graph) #获取线程坐标 coord = tf.train.Coordinator() #将队列中的多用sunner开始执行 tf.train.start_queue_runners(coord=coord, sess=sess) with sess.as_default(): if args.pretrained_model: print('Restoring pretrained model: %s' % args.pretrained_model) saver.restore(sess, os.path.expanduser(args.pretrained_model)) # Training and validation loop epoch = 0 #将所有数据过一遍的次数 默认500 while epoch < args.max_nrof_epochs: step = sess.run(global_step, feed_dict=None) #epoch_size是一个epoch中批的个数,这个epoch是全局的批处理个数以一个epoch中。。。这个epoch将用于求学习率 epoch = step // args.epoch_size # Train for one epoch train(args, sess, train_set, epoch, image_paths_placeholder, labels_placeholder, labels_batch, batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, input_queue, global_step, embeddings, total_loss, train_op, summary_op, summary_writer, args.learning_rate_schedule_file, args.embedding_size, anchor, positive, negative, triplet_loss) # Save variables and the metagraph if it doesn't exist already save_variables_and_metagraph(sess, saver, summary_writer, model_dir, subdir, step) # Evaluate on LFW if args.lfw_dir: evaluate(sess, lfw_paths, embeddings, labels_batch, image_paths_placeholder, labels_placeholder, batch_size_placeholder, learning_rate_placeholder, phase_train_placeholder, enqueue_op, actual_issame, args.batch_size, args.lfw_nrof_folds, log_dir, step, summary_writer, args.embedding_size) return model_dir
def main(args): network = importlib.import_module(args.model_def) subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S') log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir) if not os.path.isdir( log_dir): # Create the log directory if it doesn't exist os.makedirs(log_dir) model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir) if not os.path.isdir( model_dir): # Create the model directory if it doesn't exist os.makedirs(model_dir) # Write arguments to a text file facenet.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt')) # Store some git revision info in a text file in the log directory src_path, _ = os.path.split(os.path.realpath(__file__)) facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv)) np.random.seed(seed=args.seed) random.seed(args.seed) train_set = facenet.get_dataset(args.data_dir) if args.filter_filename: train_set = filter_dataset(train_set, os.path.expanduser(args.filter_filename), args.filter_percentile, args.filter_min_nrof_images_per_class) nrof_classes = len(train_set) print('Model directory: %s' % model_dir) print('Log directory: %s' % log_dir) pretrained_model = None if args.pretrained_model: pretrained_model = os.path.expanduser(args.pretrained_model) print('Pre-trained model: %s' % pretrained_model) if args.lfw_dir: print('LFW directory: %s' % args.lfw_dir) # Read the file containing the pairs used for testing pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs)) # Get the paths for the corresponding images lfw_paths, actual_issame = lfw.get_paths( os.path.expanduser(args.lfw_dir), pairs, args.lfw_file_ext) with tf.Graph().as_default(): tf.set_random_seed(args.seed) global_step = tf.Variable(0, trainable=False) # Get a list of image paths and their labels image_list, label_list = facenet.get_image_paths_and_labels(train_set) assert len(image_list) > 0, 'The dataset should not be empty' # Create a queue that produces indices into the image_list and label_list labels = ops.convert_to_tensor(label_list, dtype=tf.int32) range_size = array_ops.shape(labels)[0] index_queue = tf.train.range_input_producer(range_size, num_epochs=None, shuffle=True, seed=None, capacity=32) index_dequeue_op = index_queue.dequeue_many( args.batch_size * args.epoch_size, 'index_dequeue') learning_rate_placeholder = tf.placeholder(tf.float32, name='learning_rate') batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size') phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train') image_paths_placeholder = tf.placeholder(tf.string, shape=(None, 1), name='image_paths') labels_placeholder = tf.placeholder(tf.int64, shape=(None, 1), name='labels') input_queue = data_flow_ops.FIFOQueue(capacity=100000, dtypes=[tf.string, tf.int64], shapes=[(1, ), (1, )], shared_name=None, name=None) enqueue_op = input_queue.enqueue_many( [image_paths_placeholder, labels_placeholder], name='enqueue_op') nrof_preprocess_threads = 4 images_and_labels = [] for _ in range(nrof_preprocess_threads): filenames, label = input_queue.dequeue() images = [] for filename in tf.unstack(filenames): file_contents = tf.read_file(filename) image = tf.image.decode_image(file_contents, channels=3) if args.random_rotate: image = tf.py_func(facenet.random_rotate_image, [image], tf.uint8) if args.random_crop: image = tf.random_crop( image, [args.image_size, args.image_size, 3]) else: image = tf.image.resize_image_with_crop_or_pad( image, args.image_size, args.image_size) if args.random_flip: image = tf.image.random_flip_left_right(image) #pylint: disable=no-member image.set_shape((args.image_size, args.image_size, 3)) images.append(tf.image.per_image_standardization(image)) images_and_labels.append([images, label]) image_batch, label_batch = tf.train.batch_join( images_and_labels, batch_size=batch_size_placeholder, shapes=[(args.image_size, args.image_size, 3), ()], enqueue_many=True, capacity=4 * nrof_preprocess_threads * args.batch_size, allow_smaller_final_batch=True) image_batch = tf.identity(image_batch, 'image_batch') image_batch = tf.identity(image_batch, 'input') label_batch = tf.identity(label_batch, 'label_batch') print('Total number of classes: %d' % nrof_classes) print('Total number of examples: %d' % len(image_list)) print('Building training graph') # Build the inference graph prelogits, _ = network.inference( image_batch, args.keep_probability, phase_train=phase_train_placeholder, bottleneck_layer_size=args.embedding_size, weight_decay=args.weight_decay) logits = slim.fully_connected( prelogits, len(train_set), activation_fn=None, weights_initializer=tf.truncated_normal_initializer(stddev=0.1), weights_regularizer=slim.l2_regularizer(args.weight_decay), scope='Logits', reuse=False) embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings') # Add center loss if args.center_loss_factor > 0.0: prelogits_center_loss, _ = facenet.center_loss( prelogits, label_batch, args.center_loss_alfa, nrof_classes) tf.add_to_collection( tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_center_loss * args.center_loss_factor) learning_rate = tf.train.exponential_decay( learning_rate_placeholder, global_step, args.learning_rate_decay_epochs * args.epoch_size, args.learning_rate_decay_factor, staircase=True) tf.summary.scalar('learning_rate', learning_rate) # Calculate the average cross entropy loss across the batch cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits( labels=label_batch, logits=logits, name='cross_entropy_per_example') cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy') tf.add_to_collection('losses', cross_entropy_mean) # Calculate the total losses regularization_losses = tf.get_collection( tf.GraphKeys.REGULARIZATION_LOSSES) total_loss = tf.add_n([cross_entropy_mean] + regularization_losses, name='total_loss') # Build a Graph that trains the model with one batch of examples and updates the model parameters train_op = facenet.train(total_loss, global_step, args.optimizer, learning_rate, args.moving_average_decay, tf.global_variables(), args.log_histograms) # Create a saver saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3) # Build the summary operation based on the TF collection of Summaries. summary_op = tf.summary.merge_all() # Start running operations on the Graph. gpu_options = tf.GPUOptions( per_process_gpu_memory_fraction=args.gpu_memory_fraction) sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False)) sess.run(tf.global_variables_initializer()) sess.run(tf.local_variables_initializer()) summary_writer = tf.summary.FileWriter(log_dir, sess.graph) coord = tf.train.Coordinator() tf.train.start_queue_runners(coord=coord, sess=sess) with sess.as_default(): if pretrained_model: print('Restoring pretrained model: %s' % pretrained_model) saver.restore(sess, pretrained_model) # Training and validation loop print('Running training') epoch = 0 while epoch < args.max_nrof_epochs: step = sess.run(global_step, feed_dict=None) epoch = step // args.epoch_size # Train for one epoch train(args, sess, epoch, image_list, label_list, index_dequeue_op, enqueue_op, image_paths_placeholder, labels_placeholder, learning_rate_placeholder, phase_train_placeholder, batch_size_placeholder, global_step, total_loss, train_op, summary_op, summary_writer, regularization_losses, args.learning_rate_schedule_file) # Save variables and the metagraph if it doesn't exist already save_variables_and_metagraph(sess, saver, summary_writer, model_dir, subdir, step) # Evaluate on LFW if args.lfw_dir: evaluate(sess, enqueue_op, image_paths_placeholder, labels_placeholder, phase_train_placeholder, batch_size_placeholder, embeddings, label_batch, lfw_paths, actual_issame, args.lfw_batch_size, args.lfw_nrof_folds, log_dir, step, summary_writer) return model_dir