def inference(images, pool_type, use_lrn, keep_probability, phase_train=True, weight_decay=0.0): """ Define an inference network for face recognition based on inception modules using batch normalization Args: images: The images to run inference on, dimensions batch_size x height x width x channels phase_train: True if batch normalization should operate in training mode """ conv1 = facenet.conv(images, 3, 64, 7, 7, 2, 2, 'SAME', 'conv1_7x7', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) pool1 = facenet.mpool(conv1, 3, 3, 2, 2, 'SAME', 'pool1') if use_lrn: lrn1 = tf.nn.local_response_normalization(pool1, depth_radius=5, bias=1.0, alpha=1e-4, beta=0.75) else: lrn1 = pool1 conv2 = facenet.conv(lrn1, 64, 64, 1, 1, 1, 1, 'SAME', 'conv2_1x1', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) conv3 = facenet.conv(conv2, 64, 192, 3, 3, 1, 1, 'SAME', 'conv3_3x3', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) if use_lrn: lrn2 = tf.nn.local_response_normalization(conv3, depth_radius=5, bias=1.0, alpha=1e-4, beta=0.75) else: lrn2 = conv3 pool3 = facenet.mpool(lrn2, 3, 3, 2, 2, 'SAME', 'pool3') incept3a = facenet.inception(pool3, 192, 1, 64, 96, 128, 16, 32, 3, 32, 1, 'MAX', 'incept3a', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) incept3b = facenet.inception(incept3a, 256, 1, 64, 96, 128, 32, 64, 3, 64, 1, pool_type, 'incept3b', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) incept3c = facenet.inception(incept3b, 320, 2, 0, 128, 256, 32, 64, 3, 0, 2, 'MAX', 'incept3c', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) incept4a = facenet.inception(incept3c, 640, 1, 256, 96, 192, 32, 64, 3, 128, 1, pool_type, 'incept4a', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) incept4b = facenet.inception(incept4a, 640, 1, 224, 112, 224, 32, 64, 3, 128, 1, pool_type, 'incept4b', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) incept4c = facenet.inception(incept4b, 640, 1, 192, 128, 256, 32, 64, 3, 128, 1, pool_type, 'incept4c', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) incept4d = facenet.inception(incept4c, 640, 1, 160, 144, 288, 32, 64, 3, 128, 1, pool_type, 'incept4d', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) incept4e = facenet.inception(incept4d, 640, 2, 0, 160, 256, 64, 128, 3, 0, 2, 'MAX', 'incept4e', phase_train=phase_train, use_batch_norm=True) incept5a = facenet.inception(incept4e, 1024, 1, 384, 192, 384, 0, 0, 3, 128, 1, pool_type, 'incept5a', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) incept5b = facenet.inception(incept5a, 896, 1, 384, 192, 384, 0, 0, 3, 128, 1, 'MAX', 'incept5b', phase_train=phase_train, use_batch_norm=True, weight_decay=weight_decay) pool6 = facenet.apool(incept5b, 3, 3, 1, 1, 'VALID', 'pool6') resh1 = tf.reshape(pool6, [-1, 896]) affn1 = facenet.affine(resh1, 896, 128, 'fc7', weight_decay=weight_decay) dropout = tf.nn.dropout(affn1, keep_probability) norm = tf.nn.l2_normalize(dropout, 1, 1e-10, name='embeddings') return norm
def inference(images, pool_type, use_lrn, keep_probability, phase_train=True): """ Define an inference network for face recognition based on inception modules using batch normalization Args: images: The images to run inference on, dimensions batch_size x height x width x channels phase_train: True if batch normalization should operate in training mode """ conv1 = facenet.conv(images, 3, 64, 7, 7, 2, 2, 'SAME', 'conv1_7x7', phase_train=phase_train, use_batch_norm=True) pool1 = facenet.mpool(conv1, 3, 3, 2, 2, 'SAME') if use_lrn: lrn1 = tf.nn.local_response_normalization(pool1, depth_radius=5, bias=1.0, alpha=1e-4, beta=0.75) else: lrn1 = pool1 conv2 = facenet.conv(lrn1, 64, 64, 1, 1, 1, 1, 'SAME', 'conv2_1x1', phase_train=phase_train, use_batch_norm=True) conv3 = facenet.conv(conv2, 64, 192, 3, 3, 1, 1, 'SAME', 'conv3_3x3', phase_train=phase_train, use_batch_norm=True) if use_lrn: lrn2 = tf.nn.local_response_normalization(conv3, depth_radius=5, bias=1.0, alpha=1e-4, beta=0.75) else: lrn2 = conv3 pool3 = facenet.mpool(lrn2, 3, 3, 2, 2, 'SAME') incept3a = facenet.inception(pool3, 192, 1, 64, 96, 128, 16, 32, 3, 32, 1, 'MAX', 'incept3a', phase_train=phase_train, use_batch_norm=True) incept3b = facenet.inception(incept3a, 256, 1, 64, 96, 128, 32, 64, 3, 64, 1, pool_type, 'incept3b', phase_train=phase_train, use_batch_norm=True) incept3c = facenet.inception(incept3b, 320, 2, 0, 128, 256, 32, 64, 3, 0, 2, 'MAX', 'incept3c', phase_train=phase_train, use_batch_norm=True) incept4a = facenet.inception(incept3c, 640, 1, 256, 96, 192, 32, 64, 3, 128, 1, pool_type, 'incept4a', phase_train=phase_train, use_batch_norm=True) incept4b = facenet.inception(incept4a, 640, 1, 224, 112, 224, 32, 64, 3, 128, 1, pool_type, 'incept4b', phase_train=phase_train, use_batch_norm=True) incept4c = facenet.inception(incept4b, 640, 1, 192, 128, 256, 32, 64, 3, 128, 1, pool_type, 'incept4c', phase_train=phase_train, use_batch_norm=True) incept4d = facenet.inception(incept4c, 640, 1, 160, 144, 288, 32, 64, 3, 128, 1, pool_type, 'incept4d', phase_train=phase_train, use_batch_norm=True) incept4e = facenet.inception(incept4d, 640, 2, 0, 160, 256, 64, 128, 3, 0, 2, 'MAX', 'incept4e', phase_train=phase_train, use_batch_norm=True) incept5a = facenet.inception(incept4e, 1024, 1, 384, 192, 384, 0, 0, 3, 128, 1, pool_type, 'incept5a', phase_train=phase_train, use_batch_norm=True) incept5b = facenet.inception(incept5a, 896, 1, 384, 192, 384, 0, 0, 3, 128, 1, 'MAX', 'incept5b', phase_train=phase_train, use_batch_norm=True) pool6 = facenet.apool(incept5b, 3, 3, 1, 1, 'VALID') resh1 = tf.reshape(pool6, [-1, 896]) affn1 = facenet.affine(resh1, 896, 128) dropout = tf.nn.dropout(affn1, keep_probability) norm = tf.nn.l2_normalize(dropout, 1, 1e-10, name='embeddings') return norm