def check_gt_plot2(rgb_file, city_name, city_id, tower_gt):
if city_name == 'USA_KS_Colwich_Maize':
city_name = 'Colwich_Maize'
if city_name == 'NZ_Palmerston North':
city_name = 'PalmerstonNorth'
try:
line_gt = misc_utils.load_file(
os.path.join(
r'/media/ei-edl01/data/uab_datasets/lines_v3/data/Original_Tiles',
'{}{}_GT.png'.format(city_name, city_id)))
line_gt = cv2.dilate(line_gt, np.ones((5, 5), np.uint8), iterations=3)
plt.figure(figsize=(8, 6))
rgb = misc_utils.load_file(rgb_file)
rgb = util_functions.add_mask(rgb, line_gt, [0, 255, 0], 1)
plt.imshow(rgb)
# visualize_with_connected_pairs(rgb, tower_gt, line_gt, style='r', add_fig=True)
center_points = np.array(tower_gt)
plt.scatter(center_points[:, 1],
center_points[:, 0],
c='g',
s=40,
marker='o',
alpha=1,
edgecolors='k')
plt.title('{} {}'.format(city_name, city_id))
plt.axis('off')
plt.tight_layout()
plt.show()
except:
pass
def check_gt_plot(city_name, city_id, tower_gt, line_gt):
if city_name == 'Colwich':
city_name = 'Colwich_Maize'
img_dir = r'/home/lab/Documents/bohao/data/transmission_line/raw'
rgb_file = glob(
os.path.join(img_dir, '*_{}_{}.tif'.format(city_name, city_id)))
line_gt = misc_utils.load_file(
os.path.join(
r'/media/ei-edl01/data/uab_datasets/lines/data/Original_Tiles',
'{}{}_GT.png'.format(city_name, city_id)))
line_gt = cv2.dilate(line_gt, np.ones((5, 5), np.uint8), iterations=10)
plt.figure(figsize=(8, 6))
assert len(rgb_file) == 1
rgb = misc_utils.load_file(rgb_file[0])
rgb = util_functions.add_mask(rgb, line_gt, [0, 255, 0], 1)
plt.imshow(rgb)
# visualize_with_connected_pairs(rgb, tower_gt, line_gt, style='r', add_fig=True)
center_points = np.array(tower_gt)
plt.scatter(center_points[:, 1],
center_points[:, 0],
c='g',
s=40,
marker='o',
alpha=1,
edgecolors='k')
plt.title('{} {}'.format(city_name, city_id))
plt.axis('off')
plt.tight_layout()
plt.show()
def load_data(rgb_dir, conf_dir):
img = misc_utils.load_file(os.path.join(rgb_dir, 'USA_{}_{}.tif'.format(city_list[city_id], tile_id)))
tower_file = os.path.join(rgb_dir, 'USA_{}_{}.csv'.format(city_list[city_id], tile_id))
tower_gt = read_tower_truth(tower_file)
line_file = os.path.join(rgb_dir, 'USA_{}_{}.csv'.format(city_list[city_id], tile_id))
line_gt = read_lines_truth(line_file, tower_gt)
line_conf = misc_utils.load_file(
os.path.join(conf_dir, '{}{}.png'.format(city_list[city_id].split('_')[1], tile_id)))
return img, tower_gt, line_gt, line_conf
def main():
data_dir = r'/media/ei-edl01/data/remote_sensing_data/transmission_line/New_Zealand'
city_list = ['Gisborne', 'Palmertson', 'Rotorua', 'Tauranga']
for city in city_list:
image_dir = os.path.join(data_dir, 'New_Zealand_{}'.format(city))
img_list = glob(os.path.join(image_dir, '*.tif'))
rgb_list = natsorted([a for a in img_list if 'multiclass' not in a and 'resize' not in a])
gt_list = natsorted([a for a in img_list if 'multiclass' in a])
assert len(rgb_list) == len(gt_list)
for rgb_file, gt_file in zip(rgb_list, gt_list):
print(os.path.basename(rgb_file), os.path.basename(gt_file))
rgb = misc_utils.load_file(rgb_file)
gt = misc_utils.load_file(gt_file)
visualize.compare_figures([rgb, gt], (1, 2), fig_size=(12, 5))
def load_data(dirs, city_id, tile_id):
line_gt = misc_utils.load_file(
os.path.join(
dirs['line'],
'{}{}_GT.png'.format(city_list[city_id].split('_')[1], tile_id)))
tower_gt = get_tower_truth_pred(dirs, city_id, tile_id)
return line_gt, tower_gt
def write_features(raw_dir, city_list, city_id, tile_id, model):
raw_rgb = misc_utils.load_file(
os.path.join(raw_dir, 'USA_{}_{}.tif'.format(city_list[city_id],
tile_id)))
tile_dim = raw_rgb.shape
# get tower truth
gt_list = []
csv_file_name = os.path.join(
raw_dir, 'USA_{}_{}.csv'.format(city_list[city_id], tile_id))
for label, bbox in read_polygon_csv_data(csv_file_name):
y, x = get_center_point(*bbox)
gt_list.append([y, x])
# get features
feature = []
for start, stop, online_points in read_line_csv_data(
csv_file_name, gt_list):
centers = add_point_if_not_nearby(start, stop,
[gt_list[a] for a in online_points])
for c in centers:
c = [int(a) for a in c]
region = [
c[0] - patch_size[0] // 2, c[1] - patch_size[1] // 2,
*patch_size
]
region = get_region_bounds(region, tile_dim)
img_patch = raw_rgb[region[0]:region[0] + region[2],
region[1]:region[1] + region[3], :]
feature.append(model.get_feature(img_patch))
return np.array(feature)
def load_data(model_name, ds_name, patch_size):
assert model_name in ['unet', 'deeplab']
if model_name == 'unet':
model_dir = 'UnetCrop_inria_decay_0_PS(572, 572)_BS5_EP100_LR0.0001_DS60.0_DR0.1_SFN32'
else:
model_dir = 'DeeplabV3_inria_decay_0_PS(321, 321)_BS5_EP100_LR1e-05_DS40.0_DR0.1_SFN32'
result_file = os.path.join(r'/hdd/Results/size', model_dir,
'{}_{}'.format(ds_name,
patch_size), 'result.txt')
results = misc_utils.load_file(result_file)
return parse_results(results)
def load_data(rgb_dir, task_dir, city_id, tile_id, model_name='faster_rcnn'):
img = misc_utils.load_file(
os.path.join(rgb_dir, 'USA_{}_{}.tif'.format(city_list[city_id],
tile_id)))
pred_vis_tower = misc_utils.load_file(
os.path.join(
task_dir, 'post_{}_{}_{}_pred2.npy'.format(model_name, city_id,
tile_id)))
pred_vis_lines = misc_utils.load_file(
os.path.join(
task_dir, 'post_{}_{}_{}_conn2.npy'.format(model_name, city_id,
tile_id)))
pred_topo_tower = get_tower_pred(
os.path.join(task_dir, model_name,
'USA_{}_{}.txt'.format(city_list[city_id], tile_id)))
pred_topo_lines = misc_utils.load_file(
os.path.join(
task_dir,
'{}_graph_rnn_normal_{}_{}.npy'.format(model_name, city_id,
tile_id)))
pred_combine_tower = pred_vis_tower
pred_combine_lines = misc_utils.load_file(
os.path.join(
task_dir, '{}_graph_rnn_{}_{}.npy'.format(model_name, city_id,
tile_id)))
tower_file = os.path.join(
rgb_dir, 'USA_{}_{}.csv'.format(city_list[city_id], tile_id))
tower_gt = read_tower_truth(tower_file)
line_file = os.path.join(
rgb_dir, 'USA_{}_{}.csv'.format(city_list[city_id], tile_id))
line_gt = read_lines_truth(line_file, tower_gt)
return img, pred_vis_tower, pred_vis_lines, pred_topo_tower, pred_topo_lines, \
pred_combine_tower, pred_combine_lines, tower_gt, line_gt
def get_tower_pred(csv_file):
preds = misc_utils.load_file(csv_file)
center_list, _, _ = local_maxima_suppression(preds)
return center_list
import os
import numpy as np
from rst_utils import misc_utils, metrics
truth_dir = r'/media/ei-edl01/data/uab_datasets/inria/data/Original_Tiles'
pred_dir = '/home/lab/Documents/bohao/data/temp_results'
model_name = ['CTK', 'CTKA', 'CTK+STN_A']
for mn in model_name:
iou_a = []
iou_b = []
for tile_id in range(1, 6):
tile_name = 'vienna{}_GT.tif'.format(tile_id)
gt = misc_utils.load_file(os.path.join(truth_dir, tile_name)) // 255
pred = misc_utils.load_file(os.path.join(pred_dir, mn, 'inria', 'pred', 'vienna{}.png'.format(tile_id)))
iou_a_temp, iou_b_temp = metrics.iou_metric(gt, pred, divide_flag=True)
iou_a.append(iou_a_temp)
iou_b.append(iou_b_temp)
print('{} IoU={:.3f}'.format(mn, np.sum(iou_a)/np.sum(iou_b)))
misc_utils.set_gpu(gpu)
model = build_model(input_shape, n_filters, class_num)
optm = Adam(lr)
lrate = LearningRateScheduler(step_decay)
model_ckpt = ModelCheckpoint(os.path.join(model_save_path, 'model.hdf5'), monitor='val_loss', verbose=0,
save_best_only=False, save_weights_only=False, mode='auto')
model.compile(optm, loss='mean_squared_error', metrics=['accuracy', 'binary_crossentropy'])
feature_file_name = os.path.join(task_dir, 'mlp_tower_pair_ftr_train.npy')
label_file_name = os.path.join(task_dir, 'mlp_tower_pair_lbl_train.npy')
feature_file_name_valid = os.path.join(task_dir, 'mlp_tower_pair_ftr_valid.npy')
label_file_name_valid = os.path.join(task_dir, 'mlp_tower_pair_lbl_valid.npy')
ftr_train = misc_utils.load_file(feature_file_name)
ftr_valid = misc_utils.load_file(feature_file_name_valid)
lbl_train = misc_utils.load_file(label_file_name)
lbl_train = to_categorical(lbl_train, num_classes=class_num)
lbl_valid = misc_utils.load_file(label_file_name_valid)
lbl_valid = to_categorical(lbl_valid, num_classes=class_num)
history = model.fit(ftr_train, lbl_train, epochs=epochs, batch_size=batch_size,
validation_data=(ftr_valid, lbl_valid), callbacks=[lrate, model_ckpt])
# make plots
plt.figure(figsize=(12, 5))
plt.subplot(121)
plt.plot(history.history['binary_crossentropy'], marker='.')
plt.plot(history.history['val_binary_crossentropy'], marker='.')
'''plt.plot(history.history['mean_squared_error'], marker='.')
for city_id in range(4):
graph_train, graph_valid = [], []
centers_train, centers_valid = [], []
for tile_id in range(1, tile_id_list[city_id] + 1):
centers = []
print('Evaluating city {} tile {}'.format(city_id, tile_id))
save_file_name = os.path.join(
dirs['task'], '{}_{}_cp.npy'.format(city_list[city_id],
tile_id))
# load data
line_gt, tower_gt = load_data(dirs, city_id, tile_id)
# get tower connection info
connected_pair = misc_utils.load_file(save_file_name)
n_node = len(tower_gt)
graph = np.zeros((n_node, n_node))
for cp in connected_pair:
graph[cp[0], cp[1]] = 1
graph[cp[1], cp[0]] = 1
for i in range(n_node):
centers.append(np.array(tower_gt[i]) / np.array(line_gt.shape))
if tile_id <= 3:
graph_valid.append(graph)
centers_valid.append(centers)
else:
graph_train.append(graph)
print('Processing {}: tile {}'.format(city_list[city_id], tile_id))
pb = processBlock.ValueComputeProcess(
'{}{}_fe'.format(city_id, tile_id),
task_dir,
os.path.join(task_dir,
'{}_{}_feature.npy'.format(city_id, tile_id)),
func=write_features)
feature = pb.run(force_run=False,
raw_dir=raw_dir,
city_list=city_list,
city_id=city_id,
tile_id=tile_id,
model=res50).val
raw_rgb = misc_utils.load_file(
os.path.join(
raw_dir, 'USA_{}_{}.tif'.format(city_list[city_id],
tile_id)))
tile_dim = raw_rgb.shape
# get tower truth
gt_list = []
csv_file_name = os.path.join(
raw_dir, 'USA_{}_{}.csv'.format(city_list[city_id], tile_id))
for label, bbox in read_polygon_csv_data(csv_file_name):
y, x = get_center_point(*bbox)
gt_list.append([y, x])
current_cnt = 0
connected_pair = []
all_centers = []
for start, stop, online_points in read_line_csv_data(