def images_annotations_info(maskpath):
annotations = []
images = []
mask_images_path = loadim(maskpath)
for id_number, mask_image_path in enumerate(mask_images_path, 1):
file_name = mask_image_path.split(os.path.sep)[-1][:-4] + '.jpg'
mask_image = Image.open(mask_image_path)
sub_masks = create_sub_masks(mask_image)
for color, sub_mask in sub_masks.items():
category_id = category_ids[1][color]
# ID number
image_id = id_number
annotation_id = id_number
# image shape
width, height = mask_image.size
# 'images' info
image = create_image_annotation(file_name, height, width, image_id)
images.append(image)
# 'annotations' info
annotation = create_sub_mask_annotation(sub_mask, is_crowd,
image_id, category_id,
annotation_id)
annotations.append(annotation)
print('{:.2f}% finished.'.format(
(id_number / len(mask_images_path) * 100)))
return images, annotations
def anno_export(img_dir):
image_path_list = loadim(img_dir)
annotations = []
i = 1
print("begin to load.")
for image_path in image_path_list:
annotation = images_pose_anno(image_path)
annotations.append(annotation)
if i % 100 == 0:
print("{:.2f}% finished.".format(i / len(image_path_list) * 100))
print(annotation)
i += 1
print("all images loaded.")
pose_annotations['annotations'] = annotations
# print(pose_annotations)
return pose_annotations
def load_data():
imgs_car_door = {'car_door': [], }
latitude_car_door = []
longitude_car_door = []
for w in imgs_car_door.keys():
image_paths = load_image.loadim('/home/hangwu/CyMePro/data/car_door')
for image_path in image_paths:
resized_img = load_img(image_path)
image_name = image_path.split(os.path.sep)[-1]
la_cd, lo_cd = get_car_door_pose_from_filename(image_name)
imgs_car_door[w].append(resized_img)
latitude_car_door.append(la_cd)
longitude_car_door.append(lo_cd)
return imgs_car_door['car_door'], longitude_car_door
def load_data():
imgs_car_door = {
'car_door': [],
}
latitude_car_door = []
longitude_car_door = []
counter = 0
for w in imgs_car_door.keys():
image_paths = load_image.loadim('/home/hangwu/Workspace/car_door_half')
for image_path in image_paths:
resized_img = load_img(image_path)
image_name = image_path.split(os.path.sep)[-1]
la_cd, lo_cd = get_car_door_pose_from_filename(image_name)
imgs_car_door[w].append(resized_img)
latitude_car_door.append(la_cd)
longitude_car_door.append(lo_cd)
if counter % 100 == 0:
print("loading {:.2f}%".format(counter / len(image_paths) *
100))
counter += 1
return imgs_car_door['car_door'], latitude_car_door, longitude_car_door
if px[3] !=0:
if ymin > i:
ymin = i
if xmin > j:
xmin = j
if ymax < i:
ymax = i
if xmax < j:
xmax = j
# Bounding box information for .xml
bnd_dict['folder'] = impath.split(os.path.sep)[-2]
bnd_dict['filename'] = impath.split(os.path.sep)[-1]
bnd_dict['path'] = impath
bnd_dict['width'] = cols
bnd_dict['height'] = rows
bnd_dict['depth'] = channels
bnd_dict['xmin'] = xmin
bnd_dict['ymin'] = ymin
bnd_dict['xmax'] = xmax
bnd_dict['ymax'] = ymax
# IMPORTANT: .COPY()
# bnd_position.append(bnd_dict.copy())
return bnd_dict
if __name__ == '__main__':
# test
image_path_list = load_image.loadim('/home/hangwu/Workspace/Car_Door')
for image_path in image_path_list:
bp = object_dict(image_path)
print(bp)
import os
import cv2
# EIGEN
import load_image
def cvt(imgpath):
img = cv2.imread(imgpath)
img_gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
raws, cols = img_gray.shape
for i in range(raws):
for j in range(cols):
if img_gray[i][j] != 0:
img_gray[i][j] = 255
save_name = imgpath.split(os.path.sep)[-1][:-4]
cv2.imwrite('trimaps_with_window/{}.png'.format(save_name), img_gray)
if __name__ == '__main__':
imglist = load_image.loadim('masks_with_window')
for imgfile in imglist:
cvt(imgfile)
bnd_pos['ymax'] += move_y
# test
print(bnd_pos)
# Save images
cv2.imwrite('../data/images/{}.jpg'.format(save_name), output_image)
cv2.imwrite('../data/annotations/masks/{}.png'.format(save_name), object_mask)
'''
# Display
# cv2.imshow('{}.jpg'.format(save_name), output_image)
cv2.imshow('mask', object_mask)
cv2.waitKey(0)
cv2.destroyAllWindows()
return bnd_pos
if __name__ == '__main__':
fg_list = load_image.loadim('images')
print(fg_list)
bg_list = load_image.loadim('background', 'jpg', 'Fabrik')
print(bg_list)
for fg in fg_list:
bnd_info = generate_dict.object_dict(fg)
fg = cv2.imread(fg, -1)
bg_path = random.choice(bg_list)
print(bg_path)
bg = cv2.imread(bg_path, -1)
overlap(bg, fg, bnd_info)
if ymin > i:
ymin = i
if xmin > j:
xmin = j
if ymax < i:
ymax = i
if xmax < j:
xmax = j
# Bounding box information for .xml
bnd_dict['folder'] = impath.split(os.path.sep)[-2]
bnd_dict['filename'] = impath.split(os.path.sep)[-1]
bnd_dict['path'] = impath
bnd_dict['width'] = cols
bnd_dict['height'] = rows
bnd_dict['depth'] = channels
bnd_dict['xmin'] = xmin
bnd_dict['ymin'] = ymin
bnd_dict['xmax'] = xmax
bnd_dict['ymax'] = ymax
# IMPORTANT: .COPY()
# bnd_position.append(bnd_dict.copy())
return bnd_dict
if __name__ == '__main__':
# test
image_path_list = load_image.loadim('../data/car_door')
for image_path in image_path_list:
bp = object_dict(image_path)
print(bp)
node_ymin = SubElement(node_bndbox, 'ymin')
node_ymin.text = str(bndbox['ymin'])
node_xmax = SubElement(node_bndbox, 'xmax')
node_xmax.text = str(bndbox['xmax'])
node_ymax = SubElement(node_bndbox, 'ymax')
node_ymax.text = str(bndbox['ymax'])
# format display
xml = tostring(node_root, pretty_print=True)
xml_name = bndbox['filename'][:-4]
fp = open('../data/annotations/xmls/{}.xml'.format(xml_name), 'w')
fp.write(xml.decode())
fp.close()
if __name__ == '__main__':
fg_list = load_image.loadim('../data/car_door')
bg_list = load_image.loadim('../data/background', 'jpg', 'BG')
for fg in fg_list:
# IMPORTANT: if you want to resize images, don't forget resize in generate_dict
img_scale = 0.4
bnd_info = generate_dict.object_dict(fg, img_scale)
fg = cv2.imread(fg, -1)
# resize the car door images
fg = cv2.resize(fg, (0, 0),
fx=img_scale,
fy=img_scale,
interpolation=cv2.INTER_CUBIC)
bg_path = random.choice(bg_list)
bg = cv2.imread(bg_path, -1)
mask_bw_dest_path = "/home/hangwu/Repositories/Dataset/dataset/cardoor1/annotation_all/mask_bw"
# Car Door Subcategory: 1 or 2, IMPORTANT for naming the training data
cd_subcat = 1
# Test
test = False
if test:
fg_path = '/home/hangwu/Repositories/Mask_RCNN/dataset_tools/Test_Workspace/Image_Generation/Foreground'
bg_path = '/home/hangwu/Repositories/Mask_RCNN/dataset_tools/Test_Workspace/Image_Generation/Background'
xml_dest_path = "/home/hangwu/Repositories/Mask_RCNN/dataset_tools/Test_Workspace/Image_Generation/output/xml"
image_dest_path = "/home/hangwu/Repositories/Mask_RCNN/dataset_tools/Test_Workspace/Image_Generation/output/image"
mask_dest_path = "/home/hangwu/Repositories/Mask_RCNN/dataset_tools/Test_Workspace/Image_Generation/output/mask"
mask_bw_dest_path = "/home/hangwu/Repositories/Mask_RCNN/dataset_tools/Test_Workspace/Image_Generation/output/mask_bw"
fg_list = load_image.loadim(fg_path)
# print(fg_list[1230:1250])
bg_list = load_image.loadim(bg_path, 'jpg', '0000')
# Counter
progress_show = 1
for fg_p in fg_list:
# IMPORTANT: if you want to resize images, don't forget resize in generate_dict
img_scale = 0.4
try:
bnd_info = generate_dict.object_dict(fg_p, img_scale)
fg = cv2.imread(fg_p, -1)
# resize the car door images
fg = cv2.resize(fg, (0, 0),
fx=img_scale,
fy=img_scale,
