Merge branch 'joint' into 'master'

Add a joint lenet and inception model

See merge request !56

bob/learn/tensorflow/network/JointIncResV2Simple.py

+from .InceptionResnetV2 import inception_resnet_v2_batch_norm
+from .SimpleCNN import base_architecture as simplecnn_arch
+import numpy as np
+import tensorflow as tf
+
+
+def model_fn(features, labels, mode, params, config):
+    """The model function for join face and patch PAD. The input to the model
+    is 160x160 faces."""
+
+    faces = features['data']
+    key = features['key']
+
+    # construct patches inside the model
+    ksizes = strides = [1, 28, 28, 1]
+    rates = [1, 1, 1, 1]
+    patches = tf.extract_image_patches(faces, ksizes, strides, rates, 'VALID')
+    n_blocks = int(np.prod(patches.shape[1:3]))
+    # n_blocks should be 25 for 160x160 faces
+    patches = tf.reshape(patches, [-1, n_blocks, 28, 28, 3])
+
+    # organize the parameters
+    params = params or {}
+    learning_rate = params.get('learning_rate', 1e-4)
+    apply_moving_averages = params.get('apply_moving_averages', True)
+    n_classes = params.get('n_classes', 2)
+    add_histograms = params.get('add_histograms')
+
+    simplecnn_kwargs = {
+        'kernerl_size': (3, 3),
+        'data_format': 'channels_last',
+        'add_batch_norm': True,
+    }
+    # construct simplecnn from patches
+    for i in range(n_blocks):
+        if i == 0:
+            reuse = False
+        else:
+            reuse = True
+        with tf.variable_scope('SimpleCNN', reuse=reuse):
+            net, _ = simplecnn_arch(
+                patches[:, i], mode, **simplecnn_kwargs)
+        if i == 0:
+            simplecnn_embeddings = net
+        else:
+            simplecnn_embeddings += net
+    # average the embeddings of patches
+    simplecnn_embeddings /= n_blocks
+
+    # construct inception_resnet_v2 from faces
+    incresv2_embeddings, _ = inception_resnet_v2_batch_norm(faces, mode=mode)
+
+    embeddings = tf.concat([simplecnn_embeddings, incresv2_embeddings], 1)
+
+    # Logits layer
+    logits = tf.layers.dense(inputs=embeddings, units=n_classes, name='logits')
+
+    # # restore the model from an extra_checkpoint
+    # if extra_checkpoint is not None and mode == tf.estimator.ModeKeys.TRAIN:
+    #     tf.train.init_from_checkpoint(
+    #         ckpt_dir_or_file=extra_checkpoint["checkpoint_path"],
+    #         assignment_map=extra_checkpoint["scopes"],
+    #     )
+
+    predictions = {
+        # Generate predictions (for PREDICT and EVAL mode)
+        "classes": tf.argmax(input=logits, axis=1),
+        # Add `softmax_tensor` to the graph. It is used for PREDICT and by the
+        # `logging_hook`.
+        "probabilities": tf.nn.softmax(logits, name="softmax_tensor"),
+        'key': key,
+    }
+    if mode == tf.estimator.ModeKeys.PREDICT:
+        return tf.estimator.EstimatorSpec(mode=mode, predictions=predictions)
+
+    accuracy = tf.metrics.accuracy(
+        labels=labels, predictions=predictions["classes"])
+    metrics = {'accuracy': accuracy}
+
+    global_step = tf.train.get_or_create_global_step()
+
+    # Compute the moving average of all individual losses and the total loss.
+    if apply_moving_averages and mode == tf.estimator.ModeKeys.TRAIN:
+        variable_averages = tf.train.ExponentialMovingAverage(
+            0.9999, global_step)
+        variable_averages_op = variable_averages.apply(
+            tf.trainable_variables())
+    else:
+        variable_averages_op = tf.no_op(name='noop')
+
+    if mode == tf.estimator.ModeKeys.TRAIN:
+        # for batch normalization to be updated as well:
+        update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
+    else:
+        update_ops = []
+
+    with tf.control_dependencies([variable_averages_op] + update_ops):
+
+        # Calculate Loss (for both TRAIN and EVAL modes)
+        cross_loss = tf.losses.sparse_softmax_cross_entropy(
+            logits=logits, labels=labels)
+
+        regularization_losses = tf.get_collection(
+            tf.GraphKeys.REGULARIZATION_LOSSES)
+
+        loss = tf.add_n(
+            [cross_loss] + regularization_losses, name="total_loss")
+
+        if apply_moving_averages and mode == tf.estimator.ModeKeys.TRAIN:
+            # Compute the moving average of all individual losses and the total
+            # loss.
+            loss_averages = tf.train.ExponentialMovingAverage(0.9, name='avg')
+            loss_averages_op = loss_averages.apply(
+                tf.get_collection(tf.GraphKeys.LOSSES))
+        else:
+            loss_averages_op = tf.no_op(name='noop')
+
+        if mode == tf.estimator.ModeKeys.TRAIN:
+
+            optimizer = tf.train.GradientDescentOptimizer(
+                learning_rate=learning_rate)
+            train_op = tf.group(
+                optimizer.minimize(loss, global_step=global_step),
+                variable_averages_op, loss_averages_op)
+
+            # Log accuracy and loss
+            with tf.name_scope('train_metrics'):
+                tf.summary.scalar('accuracy', accuracy[1])
+                tf.summary.scalar('cross_entropy_loss', cross_loss)
+                tf.summary.scalar('loss', loss)
+                if apply_moving_averages:
+                    for l in tf.get_collection(tf.GraphKeys.LOSSES):
+                        tf.summary.scalar(l.op.name + "_averaged",
+                                          loss_averages.average(l))
+
+            # add histograms summaries
+            if add_histograms == 'all':
+                for v in tf.all_variables():
+                    tf.summary.histogram(v.name, v)
+            elif add_histograms == 'train':
+                for v in tf.trainable_variables():
+                    tf.summary.histogram(v.name, v)
+
+        else:
+            train_op = None
+
+    return tf.estimator.EstimatorSpec(
+        mode=mode,
+        predictions=predictions,
+        loss=loss,
+        train_op=train_op,
+        eval_metric_ops=metrics)