def getIimg(image_filename, net, number_of_classes):
from tf_image_segmentation.models.fcn_32s import FCN_32s
im = Image.open(image_filename)
heatMap = np.empty((im.size[1], im.size[0], number_of_classes))
image_filename_placeholder = tf.placeholder(tf.string)
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)
if imgNum == 0 and net_reuse == 0:
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)
else:
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=True)
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')
with tf.Session() as sess:
sess.run(initializer)
saver.restore(sess, net)
image_np, pred_np, probabilities_np = sess.run(
[image_tensor, pred, probabilities], feed_dict=feed_dict_to_use)
for i in range(0, number_of_classes):
heatMap[:, :, i] = probabilities_np.squeeze()[:, :, i]
iimg = heatMap.cumsum(axis=0).cumsum(axis=1)
return iimg
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)
# if imgNum == 0:
# reuse_var = False
# else:
# reuse_var = True
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=False)
imgNum += 1
upsampled_logit_resized = tf.image.resize_images(upsampled_logit,
(720, 1280))
pred = tf.argmax(upsampled_logit, dimension=3)
probabilities = tf.nn.softmax(upsampled_logit)
initializer = tf.local_variables_initializer()
saver = tf.train.Saver()
# image = distort_randomly_image_color(image)
resized_image, resized_annotation = scale_randomly_image_with_annotation_with_fixed_size_output(image, annotation, image_train_size)
resized_annotation = tf.squeeze(resized_annotation)
image_batch, annotation_batch = tf.train.shuffle_batch( [resized_image, resized_annotation],
batch_size=1,
capacity=samples,
num_threads=2,
min_after_dequeue=70)
upsampled_logits_batch, vgg_16_variables_mapping = FCN_32s(image_batch_tensor=image_batch,
number_of_classes=number_of_classes,
is_training=True)
valid_labels_batch_tensor, valid_logits_batch_tensor = get_valid_logits_and_labels(annotation_batch_tensor=annotation_batch,
logits_batch_tensor=upsampled_logits_batch,
class_labels=class_labels)
cross_entropies = tf.nn.softmax_cross_entropy_with_logits(logits=valid_logits_batch_tensor,
labels=valid_labels_batch_tensor)
# Normalize the cross entropy -- the number of elements
# is different during each step due to mask out regions
cross_entropy_sum = tf.reduce_mean(cross_entropies)
slim = tf.contrib.slim
from tf_image_segmentation.models.fcn_32s import FCN_32s
from tf_image_segmentation.utils.pascal_voc import pascal_segmentation_lut
from tf_image_segmentation.utils.tf_records import read_tfrecord_and_decode_into_image_annotation_pair_tensors
from tf_image_segmentation.utils.inference import adapt_network_for_any_size_input
# from tf_image_segmentation.utils.visualization import visualize_segmentation_adaptive
pascal_voc_lut = pascal_segmentation_lut()
number_of_classes = 21
image_holder = tf.placeholder(tf.float32, [None, None, None, None],
name='input')
logits, _ = FCN_32s(image_batch_tensor=image_holder,
number_of_classes=number_of_classes,
is_training=False)
saver = tf.train.Saver()
with tf.Session() as sess:
saver.restore(sess, FLAGS.ckpt_dir + "/model_fcn32s_final.ckpt")
print("Export the Model...")
graph_def = sess.graph.as_graph_def()
# import ipdb
# ipdb.set_trace()
freeze_graph_def = graph_util.convert_variables_to_constants(
sess, graph_def, ["fcn_32s/prediction"])
with open(FLAGS.output_file, 'wb') as f:
num_epochs=1)
image, annotation = read_tfrecord_and_decode_into_image_annotation_pair_tensors(
filename_queue)
# Fake batch for image and annotation by adding
# leading empty axis.
image_batch_tensor = tf.expand_dims(image, axis=0)
annotation_batch_tensor = tf.expand_dims(annotation, axis=0)
# Be careful: after adaptation, network returns final labels
# and not logits
FCN_32s = adapt_network_for_any_size_input(FCN_32s, 32)
pred, fcn_32s_variables_mapping = FCN_32s(
image_batch_tensor=image_batch_tensor,
number_of_classes=number_of_classes,
is_training=False)
# Take away the masked out values from evaluation
weights = tf.to_float(tf.not_equal(annotation_batch_tensor, 255))
# Define the accuracy metric: Mean Intersection Over Union
miou, update_op = slim.metrics.streaming_mean_iou(
predictions=pred,
labels=annotation_batch_tensor,
num_classes=number_of_classes,
weights=weights)
# The op for initializing the variables.
initializer = tf.local_variables_initializer()
