Ejemplo n.º 1
0
def main(batch_size, data_shape):
    # parser = argparse.ArgumentParser(description='predict an image on imagenet')
    # parser.add_argument('--batch-size', type=int, default=1,
    #                     help='the batch size')
    # parser.add_argument('--data-shape', type=int, default=299,
    #                     help='set image\'s shape')
    # args = parser.parse_args()
    # data_shape = (3, args.data_shape, args.data_shape)
    img_data_shape = (3, data_shape, data_shape)

    #图片格式转换
    if not os.path.exists(test_prefix_path + '.rec'):
        call_im2rec.convert(test_prefix_path, test_img_root_path)

    test = mx.io.ImageRecordIter(
        path_imgrec=test_prefix_path + '.rec',
        rand_crop=False,
        rand_mirror=False,
        data_shape=img_data_shape,
        batch_size=batch_size,
    )

    # predict
    pro, data, label = model_load.predict(test, return_data=True)

    #result process
    pro_list = pro.tolist()
    right_label = [int(x) for x in label.tolist()]
    label_dict = get_label(train_prefix_path)

    right_label_name = []
    for index in right_label:
        right_label_name.append(label_dict[index])

    predict_label = [each_pro.index(max(each_pro)) for each_pro in pro_list]
    predict_label_name = []

    for index in predict_label:
        predict_label_name.append(label_dict[index])

    return pro_list, right_label, right_label_name, predict_label, predict_label_name, label_dict
Ejemplo n.º 2
0
def main(batch_size, data_shape):
    # parser = argparse.ArgumentParser(description='predict an image on imagenet')
    # parser.add_argument('--batch-size', type=int, default=1,
    #                     help='the batch size')
    # parser.add_argument('--data-shape', type=int, default=299,
    #                     help='set image\'s shape')
    # args = parser.parse_args()
    # data_shape = (3, args.data_shape, args.data_shape)
    img_data_shape = (3, data_shape, data_shape)

    #图片格式转换
    if not os.path.exists(test_prefix_path + '.rec'):
        call_im2rec.convert(test_prefix_path, test_img_root_path)

    test = mx.io.ImageRecordIter(
            path_imgrec = test_prefix_path + '.rec',
            rand_crop   = False,
            rand_mirror = False,
            data_shape  = img_data_shape,
            batch_size  = batch_size,
            )

    # predict
    pro, data, label = model_load.predict(test, return_data=True)

    #result process
    pro_list = pro.tolist()
    right_label = [int(x) for x in label.tolist()]
    label_dict = get_label(train_prefix_path)

    right_label_name = []
    for index in right_label:
        right_label_name.append(label_dict[index])

    predict_label = [each_pro.index(max(each_pro)) for each_pro in pro_list]
    predict_label_name = []

    for index in predict_label:
        predict_label_name.append(label_dict[index])

    return pro_list, right_label, right_label_name, predict_label, predict_label_name, label_dict
Ejemplo n.º 3
0
                    help='directory of the log file')
parser.add_argument('--train-dataset', type=str, default="train.rec",
                    help='train dataset name')
parser.add_argument('--val-dataset', type=str, default="val.rec",
                    help="validation dataset name")
parser.add_argument('--data-shape', type=int, default=299,
                    help='set image\'s shape')
args = parser.parse_args()

# network
import importlib
net = importlib.import_module('symbol_' + args.network).get_symbol(args.num_classes)

# data
if not os.path.exists(train_prefix_path + '.rec'):
	call_im2rec.convert(train_prefix_path, train_img_root_path)
if not os.path.exists(val_prefix_path + '.rec'):
	call_im2rec.convert(val_prefix_path, val_img_root_path)


def get_iterator(args, kv):
    data_shape = (3, args.data_shape, args.data_shape)
    train = mx.io.ImageRecordIter(
        path_imgrec = train_prefix_path + '.rec',
        data_shape  = data_shape,
        batch_size  = args.batch_size,
        rand_crop   = True,
        rand_mirror = True,
    )

    val = mx.io.ImageRecordIter(
Ejemplo n.º 4
0
                    default="val.rec",
                    help="validation dataset name")
parser.add_argument('--data-shape',
                    type=int,
                    default=299,
                    help='set image\'s shape')
args = parser.parse_args()

# network
import importlib
net = importlib.import_module('symbol_' + args.network).get_symbol(
    args.num_classes)

# data
if not os.path.exists(train_prefix_path + '.rec'):
    call_im2rec.convert(train_prefix_path, train_img_root_path)
if not os.path.exists(val_prefix_path + '.rec'):
    call_im2rec.convert(val_prefix_path, val_img_root_path)


def get_iterator(args, kv):
    data_shape = (3, args.data_shape, args.data_shape)
    train = mx.io.ImageRecordIter(
        path_imgrec=train_prefix_path + '.rec',
        data_shape=data_shape,
        batch_size=args.batch_size,
        rand_crop=True,
        rand_mirror=True,
    )

    val = mx.io.ImageRecordIter(