def visualize_data(img, line_conf, img_dir):
line_conf = gaussian_filter(line_conf, 25)
img = img[7110:7500, 6880:7420, :]
line_conf = line_conf[7110:7500, 6880:7420]
visualize_utils.compare_figures((img, line_conf), (1, 2), show_axis=False, fig_size=(24, 10), show_fig=False)
save_dir = os.path.join(img_dir, 'demo_save')
if not os.path.exists(save_dir):
os.makedirs(save_dir)
plt.savefig(os.path.join(save_dir, 'demo.png'))
def check_annotations():
data_dir = r'/home/lab/Documents/bohao/data/transmission_line/raw2'
csv_files_temp = natsorted(glob(os.path.join(data_dir, '*.csv')))
csv_files = []
for c in csv_files_temp:
if 'NZ' in c:
if 'resize' in c:
csv_files.append(c)
else:
csv_files.append(c)
rgb_files = natsorted(glob(os.path.join(data_dir, '*.tif')))
for cn, rn in zip(csv_files, rgb_files):
if 'Dunedin' in cn:
rgb = ersa_utils.load_file(rn)
h, w = rgb.shape[:2]
gt = np.zeros((h, w), dtype=np.uint8)
for label, y, x in read_polygon_csv_data(cn):
gt[y, x] = ENCODER[label]
visualize_utils.compare_figures([rgb, gt], (1, 2), fig_size=(12, 5), show_axis=False)
def clean_conf_map(pred):
kernel = np.ones((25, 25), np.uint8)
pred = cv2.morphologyEx(pred, cv2.MORPH_OPEN, np.ones((10, 10), np.uint8))
visualize_utils.compare_figures([rgb, gt, conf, pred], (2, 2), fig_size=(10, 8), show_axis=True)
pred = cv2.morphologyEx(pred, cv2.MORPH_CLOSE, kernel, iterations=2)
return pred
for city_idx in range(3, 4):#range(len(city_list)):
model_name = 'UnetCrop_lines_city{}_pw{}_0_PS(572, 572)_BS5_EP50_LR0.0001_DS30_DR0.1_SFN32'.format(city_idx, weight)
results_dir = os.path.join(r'/hdd/Results', ds_name, model_name, ds_name+'_tw1', 'pred')
conf_dir = os.path.join(task_dir, 'confmap_uab_{}'.format(model_name))
raw_data_dir = r'/home/lab/Documents/bohao/data/transmission_line/raw'
rgb_files = natsorted([a for a in glob(os.path.join(raw_data_dir, '*.tif'))
if 'multiclass' not in a])
'''gt_files = natsorted([a for a in glob(os.path.join(raw_data_dir, '*.tif'))
if 'multiclass' in a])'''
gt_files = natsorted(glob(os.path.join(r'/media/ei-edl01/data/uab_datasets/lines_tw1/data/Original_Tiles',
'*_GT.png')))
pred_files = natsorted(glob(os.path.join(conf_dir, '*.png')))
for conf_file in pred_files:
city_with_id = os.path.basename(conf_file)[:-4]
city_name = ''.join([a for a in city_with_id if not a.isdigit()])
city_id = ''.join([a for a in city_with_id if a.isdigit()])
if city_name == city_list[city_idx]:
rgb_file = [a for a in rgb_files if city_name in a and city_id in a][0]
gt_file = [a for a in gt_files if city_name in a and city_id in a][0]
pred_file = os.path.join(results_dir, '{}{}.png'.format(city_name, city_id))
rgb = ersa_utils.load_file(rgb_file)
gt = ersa_utils.load_file(gt_file)
conf = ersa_utils.load_file(conf_file)
pred = ersa_utils.load_file(pred_file)
visualize_utils.compare_figures([rgb, gt, conf, pred], (2, 2), fig_size=(10, 8), show_axis=True)
rgbs = natsorted(glob(os.path.join(data_dir, '*.tif')))
gts = natsorted(glob(os.path.join(data_dir, '*.png')))
img_dir, task_dir = sis_utils.get_task_img_folder()
save_dir = os.path.join(img_dir, 'bihar_patch_preds')
ersa_utils.make_dir_if_not_exist(save_dir)
with tf.Session() as sess:
model.load(model_dir, sess)
for rgb_file, gt_file in zip(rgbs, gts):
rgb = ersa_utils.load_file(rgb_file)
gt = ersa_utils.load_file(gt_file)
X_batch = np.expand_dims(rgb - img_mean, axis=0)
pred = sess.run(model.output,
feed_dict={
model.inputs['X']: X_batch,
model.trainable: False
})
pred = np.argmax(pred[0, :, :, :], axis=-1)
save_name = os.path.join(
save_dir, 'cmp_{}.png'.format(os.path.basename(rgb_file)[:-4]))
visualize_utils.compare_figures([rgb.astype(np.uint8), gt, pred],
(1, 3),
fig_size=(15, 5),
show_fig=False)
plt.savefig(save_name)
plt.close()
road = add_lines(road, road_reader)
rgb = down_sample(rgb, resize_factor).astype(np.uint8)
building = down_sample(building, resize_factor)
road = down_sample(road, resize_factor)
rgb_save_name = os.path.join(save_dir, '{}{}_RGB.tif'.format(city_name, city_id+1))
building_save_name = os.path.join(save_dir, '{}{}_BGT.png'.format(city_name, city_id+1))
road_save_name = os.path.join(save_dir, '{}{}_RGT.png'.format(city_name, city_id+1))
ersa_utils.save_file(rgb_save_name, rgb.astype(np.uint8))
ersa_utils.save_file(building_save_name, building.astype(np.uint8))
ersa_utils.save_file(road_save_name, road.astype(np.uint8))
if check_result:
visualize_utils.compare_figures([rgb, building, road], (1, 3), fig_size=(12, 4))
if make_uab:
blCol = uab_collectionFunctions.uabCollection(city_name)
blCol.readMetadata()
img_mean = blCol.getChannelMeans([0, 1, 2])
'''save_dir = r'/media/ei-edl01/data/uab_datasets/AIOI/data/Original_Tiles'
rgb_files = natsorted(glob(os.path.join(save_dir, '*_RGB.tif')))
bui_files = natsorted(glob(os.path.join(save_dir, '*_BGT.png')))
roa_files = natsorted(glob(os.path.join(save_dir, '*_RGT.png')))
for i, b, r in zip(rgb_files, bui_files, roa_files):
im = ersa_utils.load_file(i)
bm = ersa_utils.load_file(b)
rm = ersa_utils.load_file(r)
import os
import matplotlib.pyplot as plt
from glob import glob
from natsort import natsorted
import sis_utils
import ersa_utils
from visualize import visualize_utils
img_dir, task_dir = sis_utils.get_task_img_folder()
data_dir = r'/media/ei-edl01/data/uab_datasets/bihar_building/data/Original_Tiles'
rgb_files = natsorted(glob(os.path.join(data_dir, '*.tif')))
gt_files = natsorted(glob(os.path.join(data_dir, '*.png')))
save_dir = os.path.join(img_dir, 'annotation_cmp')
ersa_utils.make_dir_if_not_exist(save_dir)
for cnt, (rgb_file, gt_file) in enumerate(zip(rgb_files, gt_files)):
rgb = ersa_utils.load_file(rgb_file)
gt = ersa_utils.load_file(gt_file)
visualize_utils.compare_figures([rgb, gt], (1, 2),
fig_size=(12, 5),
show_fig=False)
plt.savefig(os.path.join(save_dir, '{}.png'.format(cnt)))
plt.close()