#except:
# pass
b2 = time()
if((step[1]==float("Inf") or step[1]==0) and save):
save = False
for j in range(8):
cl.visual.save(image[j], "dump/img/source/"+str(j))
for k in range(4):
cl.visual.save(step[2][j,:,:,k], "dump/img/f"+str(k)+"/"+str(j))
cl.visual.save(step[3][j,:,:,k], "dump/tmp/f"+str(k)+"/"+str(j))
cl.visual.save(template[j], "dump/tmp/template/"+str(j))
raise Exception('error')
b = time()
print('iteration',i, format(b-a, '.2f'), format(step[1], '.2f'), np.mean(label) )
except KeyboardInterrupt():
print('exiting')
finally:
cl.tf.global_session().model_save()
print('saved')
cl.tf.global_session().close_sess()
# write this
def evaluate(model, test_data, testing_steps, i):
return
if __name__ == "__main__":
hparams = cl.hparams(name="default")
train(hparams)
return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))
def _bytes_feature(value):
return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))
def convert_to(source, filename, size):
"""Converts a dataset to tfrecords."""
print('Writing', filename)
writer = tf.python_io.TFRecordWriter(filename)
for i in range(size):
t1 = time.time()
image = source.get_sequence()
t2 = time.time()
print(t2 - t1)
image_raw = np.asarray(image, dtype=np.uint8).tostring()
ex = tf.train.Example(features=tf.train.Features(
feature={'image': _bytes_feature(image_raw)}))
writer.write(ex.SerializeToString())
writer.close()
print('Saved ', size)
if __name__ == "__main__":
hparams = cl.hparams(name="preprocessing")
d = DataGenerator(hparams)
convert_to(d, hparams.tfrecord_train_dest, hparams.train_size)
convert_to(d, hparams.tfrecord_test_dest, hparams.test_size)
#FIXME this to go to hparams
s_width = 512
BATCH_SIZE = 4
size = 100
width = 160
#MODEL_DIR = 'logs/NCCNet_flyem/'
features = { "inputs":"image:0", "outputs": "output/image_0:0"}
#features = {"inputs": "image:0", "outputs": "add_96:0"}
#features = { "inputs":"image:0", "outputs": "output/image:0"}
# Init
hparams = cl.hparams(name="evaluation")
#d = Data(hparams, random=False)
#print('data loaded')
#image, template, _ = d.get_batch(switching=False) #Sample test
print('model loaded')
def get_batch():
#image, template, _ = d.get_batch(switching=False)
#template = np.pad(template, 128, mode='constant')
image = np.zeros((BATCH_SIZE,s_width,s_width,3), dtype=np.uint8)
template = np.zeros((BATCH_SIZE,s_width,s_width,3), dtype=np.uint8)
i=0