def test_closing(dir, thresh):
print('Evaluating at {}...'.format(thresh))
# data prepare step
Data = rsrClassData(flags.rsr_data_dir)
(collect_files_test,
meta_test) = Data.getCollectionByName(flags.test_data_dir)
iou = []
for (image_name, label_name) in collect_files_test:
c_names = flags.city_name.split(',')
for c_name in c_names:
if c_name in image_name:
city_name = re.findall('[a-z\-]*(?=[0-9]+\.)', image_name)[0]
tile_id = re.findall('[0-9]+(?=\.tif)', image_name)[0]
pred_labels = scipy.misc.imread(
os.path.join(dir, '{}_{}.png'.format(city_name, tile_id)))
pred_labels = img_closing(thresh, pred_labels) * 255
# evaluate
truth_label_img = scipy.misc.imread(
os.path.join(flags.rsr_data_dir, label_name))
iou.append(sis_utils.iou_metric(truth_label_img, pred_labels))
print('{}_{} th={} iou={}'.format(city_name, tile_id, thresh,
iou[-1]))
print(np.mean(iou))
def get_fusion_ious(model_dirs):
model_name = [a.split('/')[-1] for a in model_dirs]
iou_record = {}
print('Evaluating using {}...'.format('+'.join(model_name)))
# data prepare step
Data = rsrClassData(flags.rsr_data_dir)
(collect_files_test, meta_test) = Data.getCollectionByName(flags.test_data_dir)
for (image_name, label_name) in collect_files_test:
c_names = flags.city_name.split(',')
for c_name in c_names:
if c_name in image_name:
city_name = re.findall('[a-z\-]*(?=[0-9]+\.)', image_name)[0]
tile_id = re.findall('[0-9]+(?=\.tif)', image_name)[0]
preds = np.zeros((5000, 5000, 2))
for dir in model_dirs:
preds += np.load(os.path.join(dir, '{}_{}.npy'.format(city_name, tile_id)))
pred_labels = sis_utils.get_pred_labels(preds)
# evaluate
truth_label_img = scipy.misc.imread(os.path.join(flags.rsr_data_dir, label_name))
iou = sis_utils.iou_metric(truth_label_img, pred_labels * 255)
iou_record[image_name] = iou
print('{}_{}: iou={:.2f}'.format(city_name, tile_id, iou * 100))
if len(model_name) <= 3:
np.save(os.path.join(task_dir, '{}.npy').format('_'.join(model_name)), iou_record)
else:
np.save(os.path.join(task_dir, 'fusion_{}_{}.npy').format(len(model_name),
'_'.join([a[:4] for a in model_name])), iou_record)
iou_mean = []
for _, val in iou_record.items():
iou_mean.append(val)
print(np.mean(iou_mean))
return np.mean(iou_mean)
def main(flags):
# set gpu
os.environ['CUDA_DEVICE_ORDER'] = 'PCI_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = flags.GPU
# environment settings
np.random.seed(flags.random_seed)
tf.set_random_seed(flags.random_seed)
# data prepare step
Data = rsrClassData(flags.rsr_data_dir)
(collect_files_test, meta_test) = Data.getCollectionByName(flags.test_data_dir)
# image reader
coord = tf.train.Coordinator()
# define place holder
X = tf.placeholder(tf.float32, shape=[None, flags.input_size[0], flags.input_size[1], 3], name='X')
y = tf.placeholder(tf.int32, shape=[None, flags.input_size[0], flags.input_size[1], 1], name='y')
mode = tf.placeholder(tf.bool, name='mode')
# initialize model
model = unet.UnetModel({'X':X, 'Y':y}, trainable=mode, model_name=flags.model_name, input_size=flags.input_size)
model.create_graph('X', flags.num_classes)
model.make_update_ops('X', 'Y')
# set ckdir
model.make_ckdir(flags.ckdir)
# set up graph and initialize
config = tf.ConfigProto()
# run training
start_time = time.time()
with tf.Session(config=config) as sess:
init = tf.global_variables_initializer()
sess.run(init)
saver = tf.train.Saver(var_list=tf.global_variables(), max_to_keep=1)
if os.path.exists(model.ckdir) and tf.train.get_checkpoint_state(model.ckdir):
latest_check_point = tf.train.latest_checkpoint(model.ckdir)
saver.restore(sess, latest_check_point)
threads = tf.train.start_queue_runners(coord=coord, sess=sess)
try:
for (image_name, label_name) in collect_files_test:
if flags.city_name in image_name:
city_name = re.findall('[a-z\-]*(?=[0-9]+\.)', image_name)[0]
tile_id = re.findall('[0-9]+(?=\.tif)', image_name)[0]
# load reader
iterator_test = image_reader.image_label_iterator(
os.path.join(flags.rsr_data_dir, image_name),
batch_size=flags.batch_size,
tile_dim=meta_test['dim_image'][:2],
patch_size=flags.input_size,
overlap=0)
# run
result = model.test('X', sess, iterator_test)
pred_label_img = sis_utils.get_output_label(result, meta_test['dim_image'],
flags.input_size, meta_test['colormap'])
# evaluate
truth_label_img = scipy.misc.imread(os.path.join(flags.rsr_data_dir, label_name))
iou = sis_utils.iou_metric(truth_label_img, pred_label_img)
print('{}_{}: iou={:.2f}'.format(city_name, tile_id, iou*100))
finally:
coord.request_stop()
coord.join(threads)
duration = time.time() - start_time
print('duration {:.2f} minutes'.format(duration/60))
def test(flags, ep, ds, lr, save_dir):
# set gpu
os.environ['CUDA_DEVICE_ORDER'] = 'PCI_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = flags.GPU
# environment settings
np.random.seed(flags.random_seed)
tf.set_random_seed(flags.random_seed)
# data prepare step
Data = rsrClassData(flags.rsr_data_dir)
(collect_files_test, meta_test) = Data.getCollectionByName(flags.test_data_dir)
# image reader
coord = tf.train.Coordinator()
# define place holder
X = tf.placeholder(tf.float32, shape=[None, flags.input_size[0], flags.input_size[1], 3], name='X')
y = tf.placeholder(tf.int32, shape=[None, flags.input_size[0], flags.input_size[1], 1], name='y')
mode = tf.placeholder(tf.bool, name='mode')
# initialize model
flags.model_name = 'ResUnetShrinkInria_fr_resample_mean_reduced_EP-{}_DS-{}.0_LR-{}'.format(ep, ds, lr)
model = unet.ResUnetModel_shrink({'X':X, 'Y':y}, trainable=mode, model_name=flags.model_name, input_size=flags.input_size)
model.create_graph('X', flags.num_classes)
model.make_update_ops('X', 'Y')
# set ckdir
model.make_ckdir(flags.ckdir)
# set up graph and initialize
config = tf.ConfigProto()
# run training
start_time = time.time()
with tf.Session(config=config) as sess:
init = tf.global_variables_initializer()
sess.run(init)
saver = tf.train.Saver(var_list=tf.global_variables(), max_to_keep=1)
if os.path.exists(model.ckdir) and tf.train.get_checkpoint_state(model.ckdir):
latest_check_point = tf.train.latest_checkpoint(model.ckdir)
saver.restore(sess, latest_check_point)
print('loaded {}'.format(latest_check_point))
threads = tf.train.start_queue_runners(coord=coord, sess=sess)
try:
iou_record = {}
for (image_name, label_name) in collect_files_test:
c_names = flags.city_name.split(',')
for c_name in c_names:
if c_name in image_name:
city_name = re.findall('[a-z\-]*(?=[0-9]+\.)', image_name)[0]
tile_id = re.findall('[0-9]+(?=\.tif)', image_name)[0]
# load reader
iterator_test = image_reader.image_label_iterator(
os.path.join(flags.rsr_data_dir, image_name),
batch_size=flags.batch_size,
tile_dim=meta_test['dim_image'][:2],
patch_size=flags.input_size,
overlap=0, padding=0,
image_mean=IMG_MEAN)
# run
result = model.test('X', sess, iterator_test)
pred_label_img = sis_utils.get_output_label(result,
meta_test['dim_image'],
flags.input_size,
meta_test['colormap'], overlap=0,
output_image_dim=meta_test['dim_image'],
output_patch_size=flags.input_size,
make_map=False)
# evaluate
truth_label_img = scipy.misc.imread(os.path.join(flags.rsr_data_dir, label_name))
iou = sis_utils.iou_metric(truth_label_img, pred_label_img * 255)
'''plt.subplot(121)
plt.imshow(truth_label_img)
plt.subplot(122)
plt.imshow(pred_label_img)
plt.show()'''
iou_record[image_name] = iou
print('{}_{}: iou={:.2f}'.format(city_name, tile_id, iou*100))
finally:
coord.request_stop()
coord.join(threads)
duration = time.time() - start_time
print('duration {:.2f} minutes'.format(duration/60))
np.save(os.path.join(save_dir, '{}.npy').format(model.model_name), iou_record)
iou_mean = []
for _, val in iou_record.items():
iou_mean.append(val)
print('{}: {}'.format(flags.model_name, np.mean(iou_mean)))
return np.mean(iou_mean)
def test_real_across_city(flags, MODEL_NAME, patch_size):
# set gpu
os.environ['CUDA_DEVICE_ORDER'] = 'PCI_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = flags.GPU
tf.reset_default_graph()
# environment settings
np.random.seed(flags.random_seed)
tf.set_random_seed(flags.random_seed)
result_dict = {}
for city in ['austin', 'chicago', 'kitsap', 'tyrol-w', 'vienna']:
if not os.path.exists('./temp'):
os.makedirs('./temp')
flags.city_name = city
for m_name in MODEL_NAME:
tf.reset_default_graph()
# data prepare step
Data = rsrClassData(flags.rsr_data_dir)
(collect_files_test,
meta_test) = Data.getCollectionByName(flags.test_data_dir)
# image reader
coord = tf.train.Coordinator()
# define place holder
X = tf.placeholder(tf.float32,
shape=[None, patch_size[0], patch_size[1], 3],
name='X')
y = tf.placeholder(tf.int32,
shape=[None, patch_size[0], patch_size[1], 1],
name='y')
mode = tf.placeholder(tf.bool, name='mode')
# initialize model
model = unet.UnetModel_Origin({
'X': X,
'Y': y
},
trainable=mode,
model_name=m_name,
input_size=patch_size)
model.create_graph('X', flags.num_classes)
model.make_update_ops('X', 'Y')
# set ckdir
model.make_ckdir(flags.ckdir)
# set up graph and initialize
config = tf.ConfigProto()
# run training
start_time = time.time()
with tf.Session(config=config) as sess:
init = tf.global_variables_initializer()
sess.run(init)
saver = tf.train.Saver(var_list=tf.global_variables(),
max_to_keep=1)
if os.path.exists(
model.ckdir) and tf.train.get_checkpoint_state(
model.ckdir):
print('loading model from {}'.format(model.ckdir))
latest_check_point = tf.train.latest_checkpoint(
model.ckdir)
saver.restore(sess, latest_check_point)
threads = tf.train.start_queue_runners(coord=coord, sess=sess)
for (image_name, label_name) in collect_files_test:
if flags.city_name in image_name:
city_name = re.findall('[a-z\-]*(?=[0-9]+\.)',
image_name)[0]
tile_id = re.findall('[0-9]+(?=\.tif)', image_name)[0]
print('Evaluating {}_{} at patch size: {}'.format(
city_name, tile_id, patch_size))
# load reader
iterator_test = image_reader.image_label_iterator(
os.path.join(flags.rsr_data_dir, image_name),
batch_size=flags.batch_size,
tile_dim=meta_test['dim_image'][:2],
patch_size=patch_size,
overlap=184,
padding=92)
# run
result = model.test('X', sess, iterator_test)
raw_pred = patch_extractor.un_patchify_shrink(
result, (meta_test['dim_image'][0] + 184,
meta_test['dim_image'][1] + 184),
tile_dim_output=meta_test['dim_image'],
patch_size=patch_size,
patch_size_output=(patch_size[0] - 184,
patch_size[1] - 184),
overlap=184)
file_name = '{}_{}_{}.npy'.format(
m_name.split('/')[-1], city_name, tile_id)
np.save(os.path.join('./temp', file_name), raw_pred)
coord.request_stop()
coord.join(threads)
duration = time.time() - start_time
print('duration {:.2f} minutes'.format(duration / 60))
for tile_num in range(5):
output = np.zeros((5000, 5000, 2))
for m_name in MODEL_NAME:
raw_pred = np.load(
os.path.join(
'./temp', '{}_{}_{}.npy'.format(
m_name.split('/')[-1], city, tile_num + 1)))
output += raw_pred
# combine results
pred_label_img = sis_utils.get_pred_labels(output)
output_pred = sis_utils.make_output_file(pred_label_img,
meta_test['colormap'])
# evaluate
truth_label_img = scipy.misc.imread(
os.path.join(flags.rsr_data_dir, 'inria', 'truth',
'{}{}.tif'.format(city, tile_num + 1)))
iou = sis_utils.iou_metric(truth_label_img, output_pred)
print('{}_{}: iou={:.2f}'.format(city, tile_id, iou * 100))
result_dict['{}{}'.format(city, tile_id)] = iou
shutil.rmtree('./temp')
return result_dict
