capacity=min_queue_examples + (num_threads+2)*BATCH_SIZE,
seed=3,
min_after_dequeue=min_queue_examples)
return input_batch, mask_batch, label_batch
def next_batch(inputrgb_queue, inputmsk_queue, label_queue):
input_batch, mask_batch, label_batch = train_label_reader(inputrgb_queue, inputmsk_queue, label_queue)
batch = tf.concat([input_batch, mask_batch, label_batch], axis=3)
return batch
sess = tf.InteractiveSession()
images = tf.placeholder("float")
batch_images = tf.expand_dims(images, 0)
vgg_fcn = fcn8_vgg_ours.FCN8VGG()
with tf.name_scope("content_vgg"):
vgg_fcn.build(batch_images, debug=True)
labels = tf.placeholder("int32", [None, HEIGHT, WIDTH])
loss = fcn8_vgg_ours.pixel_wise_cross_entropy(vgg_fcn.upscore32, labels, num_classes = 2)
tf.summary.scalar('pixel_wise_cross_entropy_loss', loss)
merged_summary = tf.summary.merge_all()
train_writer = tf.summary.FileWriter('./train_summary', sess.graph)
train_step = tf.train.AdamOptimizer(1e-4,0.9).minimize(loss)
logging.info("********* CNN constructed *********")
train_file = "./all_train_imgs.csv"
stream=sys.stdout)
from tensorflow.python.framework import ops
img1 = scp.misc.imread("./test_data/tabby_cat.png")
fake_mask = scp.misc.imread("./test_data/tabby_cat.png", 'L')
msk_layer = np.expand_dims(fake_mask, axis=2)
print(msk_layer.shape, img1.shape)
input_img = np.append(img1, msk_layer, 2)
print("shape", input_img.shape)
with tf.Session() as sess:
images = tf.placeholder("float")
feed_dict = {images: input_img}
batch_images = tf.expand_dims(images, 0)
vgg_fcn = fcn8_vgg.FCN8VGG()
with tf.name_scope("content_vgg"):
vgg_fcn.build(batch_images, debug=True)
print('Finished building Network.')
logging.warning("Score weights are initialized random.")
logging.warning("Do not expect meaningful results.")
logging.info("Start Initializing Variabels.")
init = tf.global_variables_initializer()
sess.run(init)
print('Running the Network')
tensors = [vgg_fcn.pred, vgg_fcn.pred_up]
fake_mask = scp.misc.imread("./test_data/tabby_cat.png",
'L') # ‘L’ (8-bit pixels, black and white) # &&
msk_layer = np.expand_dims(fake_mask, axis=2)
print(msk_layer.shape, img1.shape)
input_img = np.append(img1, msk_layer, 2)
print("shape", input_img.shape)
with tf.Session() as sess: # open session
images = tf.placeholder("float")
feed_dict = {
images: input_img
} # (feed_dict is just an argument for sess.run() function) # &&
batch_images = tf.expand_dims(images, 0) # (to build vgg_fcn)
vgg_fcn = fcn8_vgg.FCN8VGG(
) # build the vgg_fcn ("VGG OBJECT") for finetuning (FCN8 VGG here)
with tf.name_scope("content_vgg"):
vgg_fcn.build(
batch_images,
debug=True) # "" batch_images to build vgg_fcn ("VGG OBJECT")
print('Finished building Network.')
logging.warning("Score weights are initialized random.")
logging.warning("Do not expect meaningful results.")
logging.info("Start Initializing Variabels.")
init = tf.global_variables_initializer()
sess.run(init)