mindspore.ops.crop_and_resize

mindspore.ops.crop_and_resize(image, boxes, box_indices, crop_size, method='bilinear', extrapolation_value=0.0)

Extracts crops from the input image Tensor and resizes them.

Note

In case that the output shape depends on crop_size, the crop_size must be constant. For now, the backward of the operator only supports bilinear method, for other methods, will return 0.

Parameters

• image (Tensor) – A 4-D Tensor representing a batch of images. It has shape \((batch, image\_height, image\_width, depth)\).

• boxes (Tensor) – A 2-D Tensor with shape \((num\_boxes, 4)\) representing the normalized coordinates of the boxes to be cropped. The coordinates are specified in the form \([y1, x1, y2, x2]\), where \((y1, x1)\) is the first corner and \((y2, x2)\) is the second corner of the box. If \(y1 > y2\), the sampled crop is inverted upside down, the width dimensionis treated similarly when \(x1 > x2\). If normalized coordinates are not in range \([0, 1]\), extrapolated input image values are used instead. Supported data type: float32.

• box_indices (Tensor) – A 1-D Tensor of shape \((num\_boxes)\) representing the batch index for each box. Supported type: int32.

• crop_size (Tuple[int]) – A tuple of two elements: (crop_height, crop_width), representing the output size of the cropped and resized images. Only positive values are supported. Supported type: int32.

• method (str, optional) –

  An optional string that specifies the sampling method for resizing. It can be "bilinear" , "nearest" or "bilinear_v2" . Default: "bilinear" .

  • "nearest": Nearest neighbor interpolation. Each output pixel is assigned the value of the nearest input pixel. This method is simple and fast but can result in blocky or pixelated outputs.

  • "bilinear": Bilinear interpolation. Each output pixel is a weighted average of the four nearest input pixels, computed using bilinear interpolation. This method produces smoother results compared to nearest neighbor interpolation.

  • "bilinear_v2": The optimized variant of "bilinear", it may achieve better result(higher precision and speed) in some cases.

• extrapolation_value (float, optional) – An optional float value used extrapolation, if applicable. Default: 0.0 .

Returns

A 4-D tensor of shape \((num\_boxes, crop\_height, crop\_width, depth)\) with type(float32).

Raises

• TypeError – If image or boxes or box_indices is not a Tensor.

• TypeError – If crop_size is not a Tuple with two int32 elements.

• TypeError – If dtype of boxes is not float or that of box_indices is not int32.

• TypeError – If method is not a str.

• TypeError – If extrapolation_value is not a float.

• ValueError – If the shape rank of image is not 4.

• ValueError – If the shape rank of boxes is not 2.

• ValueError – If the second dim of boxes is not 4.

• ValueError – If the shape rank of box_indices is not 1.

• ValueError – If the first dim of box_indices is not equal to that of boxes.

• ValueError – If existing element in box_indices is out of range [0, batch).

• ValueError – If the data of crop_size is not positive.

• ValueError – If method is not one of 'bilinear', 'nearest', 'bilinear_v2'.

Supported Platforms:

Ascend GPU CPU

Examples

>>> import numpy as np
>>> from mindspore import ops, Tensor
>>> BATCH_SIZE = 1
>>> NUM_BOXES = 5
>>> IMAGE_HEIGHT = 256
>>> IMAGE_WIDTH = 256
>>> CHANNELS = 3
>>> image = np.random.normal(size=[BATCH_SIZE, IMAGE_HEIGHT, IMAGE_WIDTH, CHANNELS]).astype(np.float32)
>>> boxes = np.random.uniform(size=[NUM_BOXES, 4]).astype(np.float32)
>>> box_indices = np.random.uniform(size=[NUM_BOXES], low=0, high=BATCH_SIZE).astype(np.int32)
>>> crop_size = (24, 24)
>>> output = ops.crop_and_resize(Tensor(image), Tensor(boxes), Tensor(box_indices), crop_size)
>>> print(output.shape)
 (5, 24, 24, 3)