DepthToSpace

Description

Rearranges data from depth to spatial data blocks, which is the reverse transformation of SpaceToDepth.

Input

input: input tensor with shape [N, C, H, W], where N indicates the batch processing axis, C indicates the channel or depth, H indicates the height, and W indicates the width. The data type can be float16, float32, double, int32, uint8, int16, int8, int64, uint16, uint32, bfloat16, or uint64.

Attribute

• blocksize: The data type is int. The value is greater than or equal to 2. It specifies the size of the block to be moved.
• mode: The data type is string, indicating whether the arrangement is depth-column-row or column-row-depth. The default value is DCR.
• data_format (optional): specifies the data format. The default value is NHWC.

Output

output: output tensor with shape [N, C/(blocksize * blocksize), H * blocksize, W * blocksize]. The data type is the same as the input.

Constraints

None.

ONNX Opset Support

Opset v8/v9/v10/v11/v12/v13/v14/v15/v16/v17/v18