checkpoints_dir = '/tensorflow-FCN-textNonText/checkpoints'
number_of_classes = 38
slim = tf.contrib.slim
checkpoint_path = os.path.join(checkpoints_dir, 'vgg_16.ckpt')
from fcn_32s import FCN_32s, extract_vgg_16_mapping_without_fc8
'''---------------------------------------------------------------------------------------------------------------------
Process batch with FCN_32s
---------------------------------------------------------------------------------------------------------------------'''
is_training = tf.placeholder(tf.bool)
upsampled_logits_batch, vgg_16_variables_mapping = FCN_32s(image_batch_tensor=img_batch,
number_of_classes=number_of_classes,
is_training=is_training)
'''---------------------------------------------------------------------------------------------------------------------
Give a range to logits
---------------------------------------------------------------------------------------------------------------------'''
margin_logits = tf.constant(10.0)
upsampled_logits_batch = tf.maximum(tf.minimum(upsampled_logits_batch, margin_logits), -margin_logits)
'''---------------------------------------------------------------------------------------------------------------------
Mask the output of the network and compute loss
---------------------------------------------------------------------------------------------------------------------'''
# The FCN_32s output is a multiplication of 32. So, it should be resized as the img_batch before computing loss
upsampled_logits_batch_resized = tf.image.resize_images(upsampled_logits_batch, (720, 1280))
# upsampled_logits_batch_resized = upsampled_logits_batch
feed_dict_to_use = {image_filename_placeholder: image_filename}
image_tensor = tf.read_file(image_filename_placeholder)
image_tensor = tf.image.decode_jpeg(image_tensor, channels=3)
# Fake batch for image and annotation by adding leading empty axis.
image_batch_tensor = tf.expand_dims(image_tensor, axis=0)
# Be careful: after adaptation, network returns final labels and not logits
FCN_32s = adapt_network_for_any_size_input(FCN_32s, 32)
upsampled_logit, fcn_16s_variables_mapping_train = FCN_32s(
image_batch_tensor=image_batch_tensor,
number_of_classes=number_of_classes,
is_training=False,
reuse=None)
pred = tf.argmax(upsampled_logit, dimension=3)
probabilities = tf.nn.softmax(upsampled_logit)
initializer = tf.local_variables_initializer()
saver = tf.train.Saver()
cmap = plt.get_cmap('bwr')
#sess=tf.Session()
with tf.Session() as sess: