Python cv2.INTER_CUBIC Examples

def format_img(img, C):
	img_min_side = float(C.im_size)
	(height,width,_) = img.shape
	
	if width <= height:
		f = img_min_side/width
		new_height = int(f * height)
		new_width = int(img_min_side)
	else:
		f = img_min_side/height
		new_width = int(f * width)
		new_height = int(img_min_side)
	fx = width/float(new_width)
	fy = height/float(new_height)
	img = cv2.resize(img, (new_width, new_height), interpolation=cv2.INTER_CUBIC)
	img = img[:, :, (2, 1, 0)]
	img = img.astype(np.float32)
	img[:, :, 0] -= C.img_channel_mean[0]
	img[:, :, 1] -= C.img_channel_mean[1]
	img[:, :, 2] -= C.img_channel_mean[2]
	img /= C.img_scaling_factor
	img = np.transpose(img, (2, 0, 1))
	img = np.expand_dims(img, axis=0)
	return img, fx, fy 

def _resize_cv2(img, size, interpolation):
    img = img.transpose((1, 2, 0))
    if interpolation == PIL.Image.NEAREST:
        cv_interpolation = cv2.INTER_NEAREST
    elif interpolation == PIL.Image.BILINEAR:
        cv_interpolation = cv2.INTER_LINEAR
    elif interpolation == PIL.Image.BICUBIC:
        cv_interpolation = cv2.INTER_CUBIC
    elif interpolation == PIL.Image.LANCZOS:
        cv_interpolation = cv2.INTER_LANCZOS4
    H, W = size
    img = cv2.resize(img, dsize=(W, H), interpolation=cv_interpolation)

    # If input is a grayscale image, cv2 returns a two-dimentional array.
    if len(img.shape) == 2:
        img = img[:, :, np.newaxis]
    return img.transpose((2, 0, 1)) 

def transform_image(image,ang_range,shear_range,trans_range):

    # Rotation

    ang_rot = np.random.uniform(ang_range)-ang_range/2
    rows,cols,ch = image.shape    
    Rot_M = cv2.getRotationMatrix2D((cols/2,rows/2),ang_rot,1)

    # Translation
    tr_x = trans_range*np.random.uniform()-trans_range/2
    tr_y = trans_range*np.random.uniform()-trans_range/2
    Trans_M = np.float32([[1,0,tr_x],[0,1,tr_y]])

    # Shear
    pts1 = np.float32([[5,5],[20,5],[5,20]])

    pt1 = 5+shear_range*np.random.uniform()-shear_range/2
    pt2 = 20+shear_range*np.random.uniform()-shear_range/2

    pts2 = np.float32([[pt1,5],[pt2,pt1],[5,pt2]])

    shear_M = cv2.getAffineTransform(pts1,pts2)

    image = cv2.warpAffine(image,Rot_M,(cols,rows))
    image = cv2.warpAffine(image,Trans_M,(cols,rows))
    image = cv2.warpAffine(image,shear_M,(cols,rows))

    image = pre_process_image(image.astype(np.uint8))

    #image = cv2.cvtColor(image, cv2.COLOR_BGR2YUV)
    #image = image[:,:,0]
    #image = cv2.resize(image, (img_resize,img_resize),interpolation = cv2.INTER_CUBIC)

    return image 

def _augment(self, img, _):
        h, w = img.shape[:2]
        area = h * w
        for _ in range(10):
            targetArea = self.rng.uniform(self.crop_area_fraction, 1.0) * area
            aspectR = self.rng.uniform(self.aspect_ratio_low, self.aspect_ratio_high)
            ww = int(np.sqrt(targetArea * aspectR) + 0.5)
            hh = int(np.sqrt(targetArea / aspectR) + 0.5)
            if self.rng.uniform() < 0.5:
                ww, hh = hh, ww
            if hh <= h and ww <= w:
                x1 = 0 if w == ww else self.rng.randint(0, w - ww)
                y1 = 0 if h == hh else self.rng.randint(0, h - hh)
                out = img[y1:y1 + hh, x1:x1 + ww]
                out = cv2.resize(out, (self.target_shape, self.target_shape), interpolation=cv2.INTER_CUBIC)
                return out
        out = imgaug.ResizeShortestEdge(self.target_shape, interp=cv2.INTER_CUBIC).augment(img)
        out = imgaug.CenterCrop(self.target_shape).augment(out)
        return out 

def get_data(name, batch):
    isTrain = name == 'train'
    image_shape = 224

    if isTrain:
        augmentors = [
            # use lighter augs if model is too small
            GoogleNetResize(crop_area_fraction=0.49 if args.width_ratio < 1 else 0.08,
                           target_shape=image_shape),
            imgaug.RandomOrderAug(
                [imgaug.BrightnessScale((0.6, 1.4), clip=False),
                 imgaug.Contrast((0.6, 1.4), clip=False),
                 imgaug.Saturation(0.4, rgb=False),
                ]),
            imgaug.Flip(horiz=True),
        ]
    else:
        augmentors = [
            imgaug.ResizeShortestEdge(int(image_shape*256/224), cv2.INTER_CUBIC),
            imgaug.CenterCrop((image_shape, image_shape)),
        ]
    return get_imagenet_dataflow(args.data_dir, name, batch, augmentors, 
                       meta_dir = args.meta_dir) 

def resized_crop(img, i, j, h, w, size, interpolation=cv2.INTER_LINEAR):
    """Crop the given numpy ndarray and resize it to desired size.
    Notably used in :class:`~torchvision.transforms.RandomResizedCrop`.
    Args:
        img (numpy ndarray): Image to be cropped.
        i: Upper pixel coordinate.
        j: Left pixel coordinate.
        h: Height of the cropped image.
        w: Width of the cropped image.
        size (sequence or int): Desired output size. Same semantics as ``scale``.
        interpolation (int, optional): Desired interpolation. Default is
            ``cv2.INTER_CUBIC``.
    Returns:
        PIL Image: Cropped image.
    """
    assert _is_numpy_image(img), 'img should be numpy image'
    img = crop(img, i, j, h, w)
    img = resize(img, size, interpolation=interpolation)
    return img 

def format_img(self, img, box_2d):

        # Should this happen? or does normalize take care of it. YOLO doesnt like
        # img=img.astype(np.float) / 255

        # torch transforms
        normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                                std=[0.229, 0.224, 0.225])
        process = transforms.Compose ([
            transforms.ToTensor(),
            normalize
        ])

        # crop image
        pt1 = box_2d[0]
        pt2 = box_2d[1]
        crop = img[pt1[1]:pt2[1]+1, pt1[0]:pt2[0]+1]
        crop = cv2.resize(src = crop, dsize=(224, 224), interpolation=cv2.INTER_CUBIC)

        # recolor, reformat
        batch = process(crop)

        return batch 

def decode_pose(img_orig, heatmaps, pafs):
    param = {'thre1': 0.1, 'thre2': 0.05, 'thre3': 0.5}

    # Bottom-up approach:
    # Step 1: find all joints in the image (organized by joint type: [0]=nose, [1]=neck...)
    joint_list_per_joint_type = NMS(param, heatmaps, img_orig.shape[0] / float(heatmaps.shape[0]))
    # joint_list is an unravel'd version of joint_list_per_joint, where we add
    # a 5th column to indicate the joint_type (0=nose, 1=neck...)
    joint_list = np.array([tuple(peak) + (joint_type,) for joint_type, joint_peaks in enumerate(joint_list_per_joint_type) for peak in joint_peaks])

    # Step 2: find which joints go together to form limbs (which wrists go with which elbows)
    paf_upsamp = cv2.resize(pafs, (img_orig.shape[1], img_orig.shape[0]), interpolation=cv2.INTER_CUBIC)
    connected_limbs = find_connected_joints(param, paf_upsamp, joint_list_per_joint_type)

    # Step 3: associate limbs that belong to the same person
    person_to_joint_assoc = group_limbs_of_same_person(connected_limbs, joint_list)

    # (Step 4): plot results
    to_plot, canvas = plot_pose(img_orig, joint_list, person_to_joint_assoc)

    return to_plot, canvas, joint_list, person_to_joint_assoc 

def get_img(self):
        while True:
            img_name = self.image_files[self.index]
            label_name = img_name.replace('.jpg', '.png')
            img = cv2.imread(img_name)
            if img is None:
                print("load img failed:", img_name)
                self.next_img()
            else:
                break
        if self.birdeye == True:
            warped_img = cv2.warpPerspective(img, self.M, (4000, 4000),flags=cv2.INTER_CUBIC)
            img   = cv2.resize(warped_img, (self.cols, self.rows), interpolation=cv2.INTER_CUBIC)
        else:
            img   = cv2.resize(img, (self.cols, self.rows), interpolation=cv2.INTER_CUBIC)
        img   = img.transpose((2,0,1))
        return img, label_name 

def update_vis(self):
        ims = self.opt_engine.get_images(self.frame_id)

        if ims is not None:
            self.ims = ims

        if self.ims is None:
            return

        ims_show = []
        n_imgs = self.ims.shape[0]
        for n in range(n_imgs):
            # im = ims[n]
            im_s = cv2.resize(self.ims[n], (self.width, self.width), interpolation=cv2.INTER_CUBIC)
            if n == self.select_id and self.topK > 1:
                t = 3  # thickness
                cv2.rectangle(im_s, (t, t), (self.width - t, self.width - t), (0, 255, 0), t)
            im_s = im_s[np.newaxis, ...]
            ims_show.append(im_s)
        if ims_show:
            ims_show = np.concatenate(ims_show, axis=0)
            g_tmp = utils.grid_vis(ims_show, self.grid_size[1], self.grid_size[0]) # (nh, nw)
            self.vis_results = g_tmp.copy()
            self.update() 

def __call__(self, sample):

        # Fixed range of scales
        sc = self.scales[random.randint(0, len(self.scales) - 1)]

        for elem in sample.keys():
            if 'fname' in elem:
                continue
            tmp = sample[elem]

            if tmp.ndim == 2:
                flagval = cv2.INTER_NEAREST
            else:
                flagval = cv2.INTER_CUBIC

            tmp = cv2.resize(tmp, None, fx=sc, fy=sc, interpolation=flagval)

            sample[elem] = tmp

        return sample 

def format_img(img, C):
	img_min_side = float(C.im_size)
	(height,width,_) = img.shape
	
	if width <= height:
		f = img_min_side/width
		new_height = int(f * height)
		new_width = int(img_min_side)
	else:
		f = img_min_side/height
		new_width = int(f * width)
		new_height = int(img_min_side)
	img = cv2.resize(img, (new_width, new_height), interpolation=cv2.INTER_CUBIC)
	img = img[:, :, (2, 1, 0)]
	img = img.astype(np.float32)
	img[:, :, 0] -= C.img_channel_mean[0]
	img[:, :, 1] -= C.img_channel_mean[1]
	img[:, :, 2] -= C.img_channel_mean[2]
	img /= C.img_scaling_factor
	img = np.transpose(img, (2, 0, 1))
	img = np.expand_dims(img, axis=0)
	return img 

def _resize_image(img):
    dst_width = CFG.ARCH.INPUT_SIZE[0]
    dst_height = CFG.ARCH.INPUT_SIZE[1]
    h_old, w_old, _ = img.shape
    height = dst_height
    width = int(w_old * height / h_old)
    if width < dst_width:
        left_padding = int((dst_width - width)/2)
        right_padding = dst_width - width - left_padding
        resized_img = cv2.resize(img, (width, height), interpolation=cv2.INTER_CUBIC)
        resized_img = cv2.copyMakeBorder(resized_img, 0, 0, left_padding, right_padding,
            cv2.BORDER_CONSTANT, value=[255, 255, 255])
    else:
        resized_img = cv2.resize(img, (dst_width, height), interpolation=cv2.INTER_CUBIC)

    return resized_img 

def one_pic_to_video(image_path, output_video_path, fps, time):
    """
    一张图片合成视频
    one_pic_to_video('./../source/1.jpeg', './../source/output.mp4', 25, 10)
    :param path: 图片文件路径
    :param output_video_path:合成视频的路径
    :param fps:帧率
    :param time:时长
    :return:
    """

    image_clip = ImageClip(image_path)
    img_width, img_height = image_clip.w, image_clip.h

    # 总共的帧数
    frame_num = (int)(fps * time)

    img_size = (int(img_width), int(img_height))

    fourcc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')

    video = cv2.VideoWriter(output_video_path, fourcc, fps, img_size)

    for index in range(frame_num):
        frame = cv2.imread(image_path)
        # 直接缩放到指定大小
        frame_suitable = cv2.resize(frame, (img_size[0], img_size[1]), interpolation=cv2.INTER_CUBIC)

        # 把图片写进视频
        # 重复写入多少次
        video.write(frame_suitable)

    # 释放资源
    video.release()

    return VideoFileClip(output_video_path) 

def fixed_resize(sample, resolution, flagval=None):

    if flagval is None:
        if ((sample == 0) | (sample == 1)).all():
            flagval = cv2.INTER_NEAREST
        else:
            flagval = cv2.INTER_CUBIC

    if isinstance(resolution, int):
        tmp = [resolution, resolution]
        tmp[np.argmax(sample.shape[:2])] = int(round(float(resolution)/np.min(sample.shape[:2])*np.max(sample.shape[:2])))
        resolution = tuple(tmp)

    if sample.ndim == 2 or (sample.ndim == 3 and sample.shape[2] == 3):
        sample = cv2.resize(sample, resolution[::-1], interpolation=flagval)
    else:
        tmp = sample
        sample = np.zeros(np.append(resolution, tmp.shape[2]), dtype=np.float32)
        for ii in range(sample.shape[2]):
            sample[:, :, ii] = cv2.resize(tmp[:, :, ii], resolution[::-1], interpolation=flagval)
    return sample 

def get_mnist_data(is_train, image_size, batchsize):
    ds = MNISTCh('train' if is_train else 'test', shuffle=True)

    if is_train:
        augs = [
            imgaug.RandomApplyAug(imgaug.RandomResize((0.8, 1.2), (0.8, 1.2)), 0.3),
            imgaug.RandomApplyAug(imgaug.RotationAndCropValid(15), 0.5),
            imgaug.RandomApplyAug(imgaug.SaltPepperNoise(white_prob=0.01, black_prob=0.01), 0.25),
            imgaug.Resize((224, 224), cv2.INTER_AREA)
        ]
        ds = AugmentImageComponent(ds, augs)
        ds = PrefetchData(ds, 128*10, multiprocessing.cpu_count())
        ds = BatchData(ds, batchsize)
        ds = PrefetchData(ds, 256, 4)
    else:
        # no augmentation, only resizing
        augs = [
            imgaug.Resize((image_size, image_size), cv2.INTER_CUBIC),
        ]
        ds = AugmentImageComponent(ds, augs)
        ds = BatchData(ds, batchsize)
        ds = PrefetchData(ds, 20, 2)
    return ds 

def __init__(self, degrees, translate=None, scale=None, shear=None, interpolation=cv2.INTER_CUBIC, fillcolor=0):
        if isinstance(degrees, numbers.Number):
            if degrees < 0:
                raise ValueError("If degrees is a single number, it must be positive.")
            self.degrees = (-degrees, degrees)
        else:
            assert isinstance(degrees, (tuple, list)) and len(degrees) == 2, \
                "degrees should be a list or tuple and it must be of length 2."
            self.degrees = degrees

        if translate is not None:
            assert isinstance(translate, (tuple, list)) and len(translate) == 2, \
                "translate should be a list or tuple and it must be of length 2."
            for t in translate:
                if not (0.0 <= t <= 1.0):
                    raise ValueError("translation values should be between 0 and 1")
        self.translate = translate

        if scale is not None:
            assert isinstance(scale, (tuple, list)) and len(scale) == 2, \
                "scale should be a list or tuple and it must be of length 2."
            for s in scale:
                if s <= 0:
                    raise ValueError("scale values should be positive")
        self.scale = scale

        if shear is not None:
            if isinstance(shear, numbers.Number):
                if shear < 0:
                    raise ValueError("If shear is a single number, it must be positive.")
                self.shear = (-shear, shear)
            else:
                assert isinstance(shear, (tuple, list)) and len(shear) == 2, \
                    "shear should be a list or tuple and it must be of length 2."
                self.shear = shear
        else:
            self.shear = shear

        # self.resample = resample
        self.interpolation = interpolation
        self.fillcolor = fillcolor 

def execute(
        self, template_object: np.ndarray, target_object: np.ndarray, *_, **__
    ) -> FindItEngineResponse:
        resp = FindItEngineResponse()

        resized_target = cv2.resize(
            target_object, template_object.shape[::-1], interpolation=cv2.INTER_CUBIC
        )
        ssim = compare_ssim(resized_target, template_object)

        resp.append("conf", self.__dict__)
        resp.append("ssim", ssim, important=True)
        resp.append("ok", True, important=True)
        return resp 

def _rotate_cv2(img, angle, expand, fill, interpolation):
    if interpolation == PIL.Image.NEAREST:
        cv_interpolation = cv2.INTER_NEAREST
    elif interpolation == PIL.Image.BILINEAR:
        cv_interpolation = cv2.INTER_LINEAR
    elif interpolation == PIL.Image.BICUBIC:
        cv_interpolation = cv2.INTER_CUBIC

    _, H, W = img.shape
    affine_mat = cv2.getRotationMatrix2D((W / 2, H / 2), angle, 1)
    # Logic borrowed from Pillow
    if expand:
        # calculate output size
        yy = []
        xx = []
        for y, x in ((0, 0), (H, 0), (H, W), (0, W)):
            yy.append(
                affine_mat[1, 0] * x + affine_mat[1, 1] * y + affine_mat[1, 2])
            xx.append(
                affine_mat[0, 0] * x + affine_mat[0, 1] * y + affine_mat[0, 2])
        out_H = int(np.ceil(max(yy)) - np.floor(min(yy)))
        out_W = int(np.ceil(max(xx)) - np.floor(min(xx)))

        affine_mat[1][2] += out_H / 2 - H / 2
        affine_mat[0][2] += out_W / 2 - W / 2
    else:
        out_H = H
        out_W = W

    img = img.transpose((1, 2, 0))
    img = cv2.warpAffine(
        img, affine_mat, (out_W, out_H), flags=cv_interpolation,
        borderValue=fill)
    if img.ndim == 2:
        img = img[:, :, None]
    img = img.transpose((2, 0, 1))
    return img 

def motion_kernel(angle, d, sz=65):
    kern = np.ones((1, d), np.float32)
    c, s = np.cos(angle), np.sin(angle)
    A = np.float32([[c, -s, 0], [s, c, 0]])
    sz2 = sz // 2
    A[:,2] = (sz2, sz2) - np.dot(A[:,:2], ((d-1)*0.5, 0))
    kern = cv2.warpAffine(kern, A, (sz, sz), flags=cv2.INTER_CUBIC)
    return kern 

def __init__(self,
                 height,
                 width,
                 interpolation_modes=[cv2.INTER_NEAREST,
                                      cv2.INTER_LINEAR,
                                      cv2.INTER_CUBIC,
                                      cv2.INTER_AREA,
                                      cv2.INTER_LANCZOS4],
                 box_filter=None,
                 labels_format={'class_id': 0, 'xmin': 1, 'ymin': 2, 'xmax': 3, 'ymax': 4}):
        '''
        Arguments:
            height (int): The desired height of the output image in pixels.
            width (int): The desired width of the output image in pixels.
            interpolation_modes (list/tuple, optional): A list/tuple of integers
                that represent valid OpenCV interpolation modes. For example,
                integers 0 through 5 are valid interpolation modes.
            box_filter (BoxFilter, optional): Only relevant if ground truth bounding boxes are given.
                A `BoxFilter` object to filter out bounding boxes that don't meet the given criteria
                after the transformation. Refer to the `BoxFilter` documentation for details. If `None`,
                the validity of the bounding boxes is not checked.
            labels_format (dict, optional): A dictionary that defines which index in the last axis of the labels
                of an image contains which bounding box coordinate. The dictionary maps at least the keywords
                'xmin', 'ymin', 'xmax', and 'ymax' to their respective indices within last axis of the labels array.
        '''
        if not (isinstance(interpolation_modes, (list, tuple))):
            raise ValueError("`interpolation_mode` must be a list or tuple.")
        self.height = height
        self.width = width
        self.interpolation_modes = interpolation_modes
        self.box_filter = box_filter
        self.labels_format = labels_format
        self.resize = Resize(height=self.height,
                             width=self.width,
                             box_filter=self.box_filter,
                             labels_format=self.labels_format) 

def find_bbox(img, img_closed):  # 寻找身份证正反面区域
    """
    根据二值化结果判定并裁剪出身份证正反面区域
    :param img: 原始RGB图片
    :param img_closed: 二值化后的图片
    :return: 身份证正反面区域
    """
    (contours, _) = cv2.findContours(img_closed.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)  # 求出框的个数
    # 这里opencv如果版本不对（4.0或以上）会报错，只需把(contours, _)改成 (_, contours, _)

    contours = sorted(contours, key=cv2.contourArea, reverse=True)  # 按照面积大小排序

    countours_res = []
    for i in range(0, len(contours)):
        area = cv2.contourArea(contours[i])  # 计算面积

        if (area <= 0.4 * img.shape[0] * img.shape[1]) and (area >= 0.05 * img.shape[0] * img.shape[1]):
            # 人为设定,身份证正反面框的大小不会超过整张图片大小的0.4,不会小于0.05(这个参数随便设置的)
            rect = cv2.minAreaRect(contours[i])  # 最小外接矩,返回值有中心点坐标,矩形宽高,倾斜角度三个参数
            box = cv2.boxPoints(rect)
            left_down, right_down, left_up, right_up = point_judge([int(rect[0][0]), int(rect[0][1])], box)
            src = np.float32([left_down, right_down, left_up, right_up])  # 这里注意必须对应

            dst = np.float32([[0, 0], [int(max(rect[1][0], rect[1][1])), 0], [0, int(min(rect[1][0], rect[1][1]))],
                              [int(max(rect[1][0], rect[1][1])),
                               int(min(rect[1][0], rect[1][1]))]])  # rect中的宽高不清楚是个怎么机制,但是对于身份证,肯定是宽大于高,因此加个判定
            m = cv2.getPerspectiveTransform(src, dst)  # 得到投影变换矩阵
            result = cv2.warpPerspective(img, m, (int(max(rect[1][0], rect[1][1])), int(min(rect[1][0], rect[1][1]))),
                                         flags=cv2.INTER_CUBIC)  # 投影变换
            countours_res.append(result)
    return countours_res  # 返回身份证区域 

def darknet_preprocess(x, target_size=None):
    # Refer to the following darkflow
    # https://github.com/thtrieu/darkflow/blob/master/darkflow/net/yolo/predict.py
    if target_size is None or target_size[0] is None or target_size[1] is None:
        y = x.copy()
    else:
        h, w = target_size
        assert cv2 is not None, 'resizing requires `cv2`.'
        y = np.zeros((len(x), h, w, x.shape[3]))
        for i in range(len(x)):
            y[i] = cv2.resize(x[i], (w, h), interpolation=cv2.INTER_CUBIC)
    y = y[:, :, :, ::-1]
    y /= 255.
    return y 

def random_warp( in_image ):
    assert in_image.shape[:2] == (256,256)

    image = in_image.copy()


    scale = 5

    range_ = numpy.linspace( 128-120, 128+120, scale )
    mapx = numpy.broadcast_to( range_, (scale,scale) )
    mapy = mapx.T

    mapx = mapx + numpy.random.normal( size=(scale,scale), scale= 6 )
    mapy = mapy + numpy.random.normal( size=(scale,scale), scale= 6 )

    interp_mapx = cv2.resize( mapx, (80,80) )[8:72,8:72].astype('float32')
    interp_mapy = cv2.resize( mapy, (80,80) )[8:72,8:72].astype('float32')

    warped_image = cv2.remap( image[:,:,:3], interp_mapx, interp_mapy, cv2.INTER_CUBIC )

    src_points = numpy.stack( [ mapx.ravel(), mapy.ravel() ], axis=-1 )
    dst_points = numpy.mgrid[0:65:16,0:65:16].T.reshape(-1,2)
    mat = umeyama( src_points, dst_points, True )[0:2]

    target_image = cv2.warpAffine( image, mat, (64,64) )

    target_mask = target_image[:,:,3].reshape((64,64,1))
    target_image = target_image[:,:,:3]


    if len(target_image.shape)>2:
      return ( warped_image, 
               target_image, 
               target_mask )
    else:
      return ( warped_image, 
               target_image ) 

def load_img(paths, grayscale=False, target_size=None, crop_size=None,
             interp=None):
    assert cv2 is not None, '`load_img` requires `cv2`.'
    if interp is None:
        interp = cv2.INTER_CUBIC
    if not isinstance(paths, list):
        paths = [paths]
    if len(paths) > 1 and (target_size is None or
                           isinstance(target_size, int)):
        raise ValueError('A tuple `target_size` should be provided '
                         'when loading multiple images.')

    def _load_img(path):
        img = cv2.imread(path)
        if target_size:
            if isinstance(target_size, int):
                hw_tuple = tuple([x * target_size // min(img.shape[:2])
                                  for x in img.shape[1::-1]])
            else:
                hw_tuple = (target_size[1], target_size[0])
            if img.shape[1::-1] != hw_tuple:
                img = cv2.resize(img, hw_tuple, interpolation=interp)
        img = img[:, :, ::-1]
        if len(img.shape) == 2:
            img = np.expand_dims(img, -1)
        return img

    if len(paths) > 1:
        imgs = np.zeros((len(paths),) + target_size + (3,), dtype=np.float32)
        for (i, path) in enumerate(paths):
            imgs[i] = _load_img(path)
    else:
        imgs = np.array([_load_img(paths[0])], dtype=np.float32)

    if crop_size is not None:
        imgs = crop(imgs, crop_size)

    return imgs 

def __init__(self, size, interpolation=cv2.INTER_CUBIC):
        assert isinstance(size, int) or (isinstance(size, collections.Iterable) and len(size) == 2)
        self.size = size
        self.interpolation = interpolation 

def __init__(self, size, scale=(0.08, 1.0), ratio=(3. / 4., 4. / 3.), interpolation=cv2.INTER_CUBIC):
        self.size = (size, size)
        self.interpolation = interpolation
        self.scale = scale
        self.ratio = ratio 

def affine(img, angle, translate, scale, shear, interpolation=cv2.INTER_LINEAR, mode=cv2.BORDER_CONSTANT, fillcolor=0):
    """Apply affine transformation on the image keeping image center invariant
    Args:
        img (numpy ndarray): numpy ndarray to be transformed.
        angle (float or int): rotation angle in degrees between -180 and 180, clockwise direction.
        translate (list or tuple of integers): horizontal and vertical translations (post-rotation translation)
        scale (float): overall scale
        shear (float): shear angle value in degrees between -180 to 180, clockwise direction.
        interpolation (``cv2.INTER_NEAREST` or ``cv2.INTER_LINEAR`` or ``cv2.INTER_AREA``, ``cv2.INTER_CUBIC``):
            An optional resampling filter.
            See `filters`_ for more information.
            If omitted, it is set to ``cv2.INTER_LINEAR``, for bilinear interpolation.
        mode (``cv2.BORDER_CONSTANT`` or ``cv2.BORDER_REPLICATE`` or ``cv2.BORDER_REFLECT`` or ``cv2.BORDER_REFLECT_101``)
            Method for filling in border regions.
            Defaults to cv2.BORDER_CONSTANT, meaning areas outside the image are filled with a value (val, default 0)
        val (int): Optional fill color for the area outside the transform in the output image. Default: 0
    """
    if not _is_numpy_image(img):
        raise TypeError('img should be numpy Image. Got {}'.format(type(img)))

    assert isinstance(translate, (tuple, list)) and len(translate) == 2, \
        "Argument translate should be a list or tuple of length 2"

    assert scale > 0.0, "Argument scale should be positive"

    output_size = img.shape[0:2]
    center = (img.shape[1] * 0.5 + 0.5, img.shape[0] * 0.5 + 0.5)
    matrix = _get_affine_matrix(center, angle, translate, scale, shear)

    if img.shape[2]==1:
        return cv2.warpAffine(img, matrix, output_size[::-1],interpolation, borderMode=mode, borderValue=fillcolor)[:,:,np.newaxis]
    else:
        return cv2.warpAffine(img, matrix, output_size[::-1],interpolation, borderMode=mode, borderValue=fillcolor) 

def compute_result(self):
        im, mask = self.uiControl.get_input()
        im_mask0 = mask > 0.0
        self.im_mask0 = im_mask0.transpose((2, 0, 1))
        im_lab = color.rgb2lab(im).transpose((2, 0, 1))
        self.im_ab0 = im_lab[1:3, :, :]

        self.model.net_forward(self.im_ab0, self.im_mask0)
        ab = self.model.output_ab.transpose((1, 2, 0))
        ab_win = cv2.resize(ab, (self.win_w, self.win_h), interpolation=cv2.INTER_CUBIC)
        pred_lab = np.concatenate((self.l_win[..., np.newaxis], ab_win), axis=2)
        pred_rgb = (np.clip(color.lab2rgb(pred_lab), 0, 1) * 255).astype('uint8')
        self.result = pred_rgb
        self.emit(SIGNAL('update_result'), self.result)
        self.update() 

def get_ilsvrc_data_alexnet(is_train, image_size, batchsize, directory):
    if is_train:
        if not directory.startswith('/'):
            ds = ILSVRCTTenthTrain(directory)
        else:
            ds = ILSVRC12(directory, 'train')
        augs = [
            imgaug.RandomApplyAug(imgaug.RandomResize((0.9, 1.2), (0.9, 1.2)), 0.7),
            imgaug.RandomApplyAug(imgaug.RotationAndCropValid(15), 0.7),
            imgaug.RandomApplyAug(imgaug.RandomChooseAug([
                imgaug.SaltPepperNoise(white_prob=0.01, black_prob=0.01),
                imgaug.RandomOrderAug([
                    imgaug.BrightnessScale((0.8, 1.2), clip=False),
                    imgaug.Contrast((0.8, 1.2), clip=False),
                    # imgaug.Saturation(0.4, rgb=True),
                ]),
            ]), 0.7),
            imgaug.Flip(horiz=True),

            imgaug.ResizeShortestEdge(256, cv2.INTER_CUBIC),
            imgaug.RandomCrop((224, 224)),
        ]
        ds = AugmentImageComponent(ds, augs)
        ds = PrefetchData(ds, 1000, multiprocessing.cpu_count())
        ds = BatchData(ds, batchsize)
        ds = PrefetchData(ds, 10, 4)
    else:
        if not directory.startswith('/'):
            ds = ILSVRCTenthValid(directory)
        else:
            ds = ILSVRC12(directory, 'val')
        ds = AugmentImageComponent(ds, [
            imgaug.ResizeShortestEdge(224, cv2.INTER_CUBIC),
            imgaug.CenterCrop((224, 224)),
        ])
        ds = PrefetchData(ds, 100, multiprocessing.cpu_count())
        ds = BatchData(ds, batchsize)

    return ds