if input_type == 'UVV':
cfg.MODEL.WEIGHTS = 'out_training/KAIST_flow_scale_UVV_0414/out_model_iter_11000.pth'
else:
cfg.MODEL.WEIGHTS = 'out_training/KAIST_flow_UVM_scale_0412/out_model_iter_22000.pth'
#cfg.MODEL.ROI_HEADS.NUM_CLASSES = 17
folder = '../../../Datasets/KAIST/test/KAIST_flow_test_sanitized/'
file_list = glob.glob(os.path.join(folder, '*.flo'))
img_folder = '../../../Datasets/KAIST/test/'
# Create predictor
predictor = DefaultPredictor(cfg)
scale = 0
for i in range(len(file_list)):
fpath = file_list[i]
flow = readFlow(fpath)
image = np.zeros((flow.shape[0], flow.shape[1], 3))
image[:, :, 0] = flow[:, :, 0]
image[:, :, 1] = flow[:, :, 1]
# UVV
if input_type == 'UVV':
image[:, :, 2] = flow[:, :, 1]
if scale == 1:
image *= 3.0
image += 128.0
image[image > 255] = 255.0
else:
image = np.abs(image) / 40.0 * 255.0
image[image > 255] = 255.0
else:
continue
# rescale score 0~100
box[4] *= 100
f.write(str(i+1)+',')
f.write(','.join(str(c) for c in box))
f.write('\n')
"""
scale = 1
for i in range(len(contents)):
fpath = contents[i].split('\n')[0]
set_num = fpath.split('/')[0]
V_num = fpath.split('/')[1]
img_num = fpath.split('/')[2]
flow_path = folder + set_num + '_' + V_num + '_' + img_num + '.flo'
flow = readFlow(flow_path)
image = np.zeros((flow.shape[0], flow.shape[1], 3))
image[:,:,0] = flow[:,:,0]
image[:,:,1] = flow[:,:,1]
# UVV
if input_type == 'UVV':
image[:,:,2] = flow[:,:,1]
if scale == 1:
image *= 3.0
image += 128.0
image[image>255] = 255.0
else:
image = np.abs(image) / 40.0 * 255.0
image[image>255] = 255.0
else:
def read_image(file_name, format=None):
"""
Read an image into the given format.
Will apply rotation and flipping if the image has such exif information.
Args:
file_name (str): image file path
format (str): one of the supported image modes in PIL, or "BGR"
Returns:
image (np.ndarray): an HWC image
"""
with PathManager.open(file_name, "rb") as f:
if format == "BGRT":
"""
# KAIST
folder = file_name.split('visible')[0]
img_name = file_name.split('visible/')[1]
path_rgb = file_name
path_thermal = folder + 'lwir/' + img_name
img_rgb = cv2.imread(path_rgb)
img_thermal = cv2.imread(path_thermal)
image = np.zeros((img_thermal.shape[0], img_thermal.shape[1], 4))
image [:,:,0:3] = img_rgb
image [:,:,3] = img_thermal[:,:,0]
"""
# FLIR
folder = file_name.split('thermal_8_bit/')[0]
img_name = file_name.split('thermal_8_bit/')[1]
img_name = img_name.split('.')[0] + '.jpg'
rgb_path = folder + 'RGB/' + img_name
#print(rgb_path)
rgb_img = cv2.imread(rgb_path)
thermal_img = cv2.imread(file_name)
#import pdb; pdb.set_trace()
rgb_img = cv2.resize(rgb_img,
(thermal_img.shape[1], thermal_img.shape[0]))
image = np.zeros((thermal_img.shape[0], thermal_img.shape[1], 4))
image[:, :, 0:3] = rgb_img
image[:, :, 3] = thermal_img[:, :, 0]
#"""
elif format == 'BGR_only':
folder = file_name.split('thermal_8_bit/')[0]
img_name = file_name.split('thermal_8_bit/')[1]
img_name = img_name.split('.')[0] + '.jpg'
rgb_path = folder + 'resized_RGB/' + img_name
image = cv2.imread(rgb_path)
elif format == 'BGRTTT': # middle fusion
"""
# KAIST
folder = file_name.split('visible')[0]
img_name = file_name.split('visible/')[1]
path_rgb = file_name
path_thermal = folder + 'lwir/' + img_name
img_rgb = cv2.imread(path_rgb)
img_thermal = cv2.imread(path_thermal)
image = np.zeros((img_thermal.shape[0], img_thermal.shape[1], 6))
image [:,:,0:3] = img_rgb
image [:,:,3:] = img_thermal
"""
# FLIR
folder = file_name.split('thermal_8_bit/')[0]
img_name = file_name.split('thermal_8_bit/')[-1]
img_name = img_name.split('.')[0] + '.jpg'
rgb_path = folder + 'RGB/' + img_name
rgb_img = cv2.imread(rgb_path)
thermal_img = cv2.imread(file_name)
rgb_img = cv2.resize(rgb_img,
(thermal_img.shape[1], thermal_img.shape[0]))
image = np.zeros((thermal_img.shape[0], thermal_img.shape[1], 6))
image[:, :, 0:3] = rgb_img
image[:, :, 3:6] = thermal_img
#"""
elif format == 'BGRTTT_perturb':
folder = file_name.split('thermal_8_bit/')[0]
img_name = file_name.split('thermal_8_bit/')[1]
img_name = img_name.split('.')[0] + '.jpg'
rgb_path = folder + 'RGB/' + img_name
rgb_img = cv2.imread(rgb_path)
import os
number = int(file_name.split('video_')[-1].split('.')[0])
#number = int(file_name.split('FLIR')[-1].split('_')[1].split('.')[0])
number -= 1
number_str = '{:05d}'.format(number)
new_file_name = file_name.split('thermal')[
0] + 'thermal_8_bit/FLIR_video_' + number_str + '.jpeg'
if os.path.exists(new_file_name):
thermal_img = cv2.imread(new_file_name)
print(new_file_name, ' RGB: ', rgb_path)
else:
thermal_img = cv2.imread(file_name)
print(file_name, ' RGB: ', rgb_path)
rgb_img = cv2.resize(rgb_img,
(thermal_img.shape[1], thermal_img.shape[0]))
"""
# Random resize
import random
ratio = random.randrange(100,121) / 100
width_new = int(640*ratio+0.5)
height_new = int(512*ratio+0.5)
rgb_img = cv2.resize(rgb_img, (width_new, height_new))
# Random crop
[height, width, _] = thermal_img.shape
diff_w = width_new - width
diff_h = height_new - height
if diff_w > 0: shift_x = random.randrange(0, diff_w)
else: shift_x = 0
if diff_h > 0: shift_y = random.randrange(0, diff_h)
else: shift_y = 0
rgb_img = rgb_img[shift_y:shift_y+height, shift_x:shift_x+width, :]
"""
#import pdb; pdb.set_trace()
image = np.zeros((thermal_img.shape[0], thermal_img.shape[1], 6))
image[:, :, 0:3] = rgb_img
image[:, :, 3:6] = thermal_img
elif format == "mid_RGB_out":
thermal_img = cv2.imread(file_name)
image = np.zeros((thermal_img.shape[0], thermal_img.shape[1], 6))
image[:, :, 3:6] = thermal_img
elif format == 'T_TCONV':
#import pdb;pdb.set_trace()
folder = file_name.split('thermal_8_bit/')[0]
img_name = file_name.split('thermal_8_bit/')[1]
img_name = img_name.split('.')[0] + '.jpeg'
t_conv_path = folder + 'thermal_convert/' + img_name
t_conv_img = cv2.imread(t_conv_path)
thermal_img = cv2.imread(file_name)
image = np.zeros((thermal_img.shape[0], thermal_img.shape[1], 2))
image[:, :, 0] = t_conv_img[:, :, 0]
image[:, :, 1] = thermal_img[:, :, 0]
elif format == 'T_TCONV_MASK':
folder = file_name.split('thermal_convert/')[0]
img_name = file_name.split('thermal_convert/')[1]
#img_name = img_name.split('.')[0] + '.jpeg'
t_conv_path = folder + 'thermal_convert/' + img_name
t_conv_img = cv2.imread(t_conv_path)
t_mask_path = folder + 'thermal_analysis/' + file_name.split(
'thermal_convert/')[1].split(".")[0] + '_mask.jpg'
mask_img = cv2.imread(t_mask_path)
thermal_img = cv2.imread(file_name)
image = np.zeros((thermal_img.shape[0], thermal_img.shape[1], 3))
image[:, :, 0] = t_conv_img[:, :, 0]
image[:, :, 1] = thermal_img[:, :, 0]
image[:, :, 2] = mask_img[:, :, 0]
elif format == 'UVV': # UV in first two channel, 0 in third channel
if 'train' in file_name:
folder = '../../../Datasets/KAIST/train/KAIST_flow_train_sanitized/'
else:
folder = '../../../Datasets/KAIST/test/KAIST_flow_test_sanitized/'
fname = file_name.split('/')[-1].split('.')[0] + '.flo'
fpath = folder + fname
flow = readFlow(fpath)
image = np.zeros((flow.shape[0], flow.shape[1], 3))
image[:, :, 0] = flow[:, :, 0]
image[:, :, 1] = flow[:, :, 1]
image[:, :, 2] = flow[:, :, 1]
image *= 4.0
#image += 128.0
image[image > 255] = 255.0
#pdb.set_trace()
"""
image = np.abs(image) / 40.0 * 255.0
image[image>255] = 255.0
"""
elif format == 'UVM': # UV + magnitude(uv)
if 'train' in file_name:
folder = '../../../Datasets/KAIST/train/KAIST_flow_train_sanitized/'
else:
folder = '../../../Datasets/KAIST/test/KAIST_flow_test_sanitized/'
fname = file_name.split('/')[-1].split('.')[0] + '.flo'
fpath = folder + fname
flow = readFlow(fpath)
flow_s = flow * flow
magnitude = np.sqrt(flow_s[:, :, 0] + flow_s[:, :, 1])
image = np.zeros((flow.shape[0], flow.shape[1], 3))
image[:, :, 0] = flow[:, :, 0]
image[:, :, 1] = flow[:, :, 1]
image[:, :, 2] = magnitude
image *= 4.0
#image += 128.0
image[image > 255] = 255.0
"""
image = np.abs(image) / 40.0 * 255.0
image[image>255] = 255.0
"""
elif format == 'BGRTUV':
if 'train' in file_name:
flow_folder = '../../../Datasets/KAIST/train/KAIST_flow_train_sanitized/'
img_folder = '../../../Datasets/KAIST/train/'
else:
flow_folder = '../../../Datasets/KAIST/test/KAIST_flow_test_sanitized/'
img_folder = '../../../Datasets/KAIST/test/'
fname = file_name.split('/')[-1].split('.')[0] + '.flo'
fpath = flow_folder + fname
flow = readFlow(fpath)
image = np.zeros((flow.shape[0], flow.shape[1], 6))
image[:, :, 4] = flow[:, :, 0]
image[:, :, 5] = flow[:, :, 1]
image *= 3
image += 128.0
image[image > 255] = 255.0
set_name = file_name.split('/')[-1].split('_')[0]
V_name = file_name.split('/')[-1].split('_')[1]
img_name = file_name.split('/')[-1].split('_')[2]
fname_bgr = img_folder + set_name + '/' + V_name + '/visible/' + img_name
fname_thr = img_folder + set_name + '/' + V_name + '/lwir/' + img_name
bgr = cv2.imread(fname_bgr)
thr = cv2.imread(fname_thr)
image[:, :, 0:3] = bgr
image[:, :, 3] = thr[:, :, 0]
else:
#import pdb; pdb.set_trace()
image = Image.open(f)
# capture and ignore this bug: https://github.com/python-pillow/Pillow/issues/3973
try:
image = ImageOps.exif_transpose(image)
except Exception:
pass
if format is not None:
# PIL only supports RGB, so convert to RGB and flip channels over below
conversion_format = format
if format == "BGR":
conversion_format = "RGB"
image = image.convert(conversion_format)
image = np.asarray(image)
if format == "BGR":
# flip channels if needed
image = image[:, :, ::-1]
# PIL squeezes out the channel dimension for "L", so make it HWC
if format == "L":
image = np.expand_dims(image, -1)
return image