for i in tqdm(range(0, len(df))):
filename = df.iloc[i]['filename']
x1 = df.iloc[i]['bbox_x1']
y1 = df.iloc[i]['bbox_y1']
x2 = df.iloc[i]['bbox_x2']
y2 = df.iloc[i]['bbox_y2']
src = Image.open(osp.join(opt.dataroot, 'cars_' + datamode, filename))
area = [x1, y1, x2, y2]
cropped_img = src.crop(area)
cropped_img.save(os.path.join(output_dir, filename))
if __name__ == '__main__':
option_parser = BaseOptions()
opt = option_parser.parse()
devkit_path = 'data/devkit/'
cars_train = 'data/cars_train/'
cars_test = 'data/cars_test/'
car_meta = loadmat(devkit_path + 'cars_meta.mat')
cars_train_annos = loadmat(devkit_path + 'cars_train_annos.mat')
cars_test_annos = loadmat(devkit_path + 'cars_test_annos.mat')
train_df = load_meta(cars_train_annos, 'train')
test_df = load_meta(cars_test_annos, 'test')
crop_image(opt, train_df, osp.join(opt.dataroot, 'cars_train' + '_cropped'), 'train')
crop_image(opt, test_df, osp.join(opt.dataroot, 'cars_test' + '_cropped'), 'test')
# parser.add_argument('--cuda', default=True, type=str2bool,
# help='Use cuda to train model')
# parser.add_argument('--voc_root', default=VOCroot, help='Location of VOC root directory')
#
# args = parser.parse_args()
#
# if not os.path.exists(args.save_folder):
# os.mkdir(args.save_folder)
#
# if args.cuda and torch.cuda.is_available():
# torch.set_default_tensor_type('torch.cuda.FloatTensor')
# else:
# torch.set_default_tensor_type('torch.FloatTensor')
options = BaseOptions()
args = options.parse()
options.setup_option()
annopath = os.path.join(args.voc_root, 'VOC2007', 'Annotations', '%s.xml')
imgpath = os.path.join(args.voc_root, 'VOC2007', 'JPEGImages', '%s.jpg')
imgsetpath = os.path.join(args.voc_root, 'VOC2007', 'ImageSets', 'Main',
'{:s}.txt')
YEAR = '2007'
devkit_path = VOCroot + 'VOC' + YEAR
dataset_mean = (104, 117, 123)
set_type = 'test'
class Timer(object):
"""A simple timer."""
def __init__(self):
self.total_time = 0.
