Im2col

English | 简体中文

Product Support Status

Product	Supported
Ascend 950PR&950DT products	√
Atlas A3 products	√
Atlas A2 products	√
Atlas 200I/500 A2 inference products	×
Atlas inference products	×
Atlas training products	×

Function Description

• Operator function: image to columns. Sliding local window data is converted into column vectors and concatenated into a large tensor. Sliding windows are extracted from a batched input tensor. Consider a batched input tensor with shape (N, C, H, W) or (C, H, W), where N is the batch dimension, C is the channel dimension, and H, W indicate the image size. This operation flattens each sliding block of kernel_size within the spatial dimensions of the input into a column (that is, the last dimension) of a 3-D output tensor of (N, C $\times \prod$(kernel_size), L) or a 2-D output tensor of (C $\times \prod$(kernel_size), L), and L is the total number of such blocks.
• Computation formula: $L={\prod }_d\lfloor \frac{spatial\_size[d]+2\times padding[d]-dilation[d]\times (kernel\_size[d]-1)-1}{stride[d]}+1\rfloor$, where spatial_size is composed of H and W of the aforementioned input tensor.

Parameter Description

Parameter Name	Input/Output/Attribute	Description	Data Type	Data Format
self	Input tensor	Input tensor with a 3-dimensional or 4-dimensional shape.	INT8, UINT8, INT16, UINT16, INT32, UINT32, INT64, UINT64, BFLOAT16, FLOAT16, FLOAT, DOUBLE, BOOL, COMPLEX32, COMPLEX64	ND
kernelSize	Input array	Kernel size. The size is 2. kernelSize[0] indicates the 'H' direction, and kernelSize[1] indicates the 'W' direction.	INT64	-
dilation	Input array	Dilation parameter. The size is 2. dilation[0] indicates the 'H' direction, and dilation[1] indicates the 'W' direction.	INT64	-
padding	Input array	Padding size. The size is 2. padding[0] indicates the H direction, and padding[1] indicates the W direction.	INT64	-
stride	Input array	Stride. The size is 2. stride[0] indicates the H direction, and stride[1] indicates the W direction.	INT64	-
out	Output tensor	Output tensor whose shape is derived from the parameters.	INT8, UINT8, INT16, UINT16, INT32, UINT32, INT64, UINT64, BFLOAT16, FLOAT16, FLOAT, DOUBLE, BOOL, COMPLEX32, COMPLEX64	ND

• Atlas A3 products and Atlas A2 products: only FLOAT, FLOAT16, and BFLOAT16 are supported.

Constraint Description

• The input tensor must be 3-dimensional or 4-dimensional.
• The size of kernelSize, dilation, padding, and stride must be 2.
• The values of kernelSize, dilation, and stride must be greater than 0.
• The value of padding cannot be less than 0.

Invocation Description

Invocation Method	Sample Code	Description
aclnn invocation	test_aclnn_im2col	Invokes the Im2col operator through the aclnnIm2col API.
Graph-mode invocation	test_geir_im2col	Invokes the Im2col operator by building a graph through the operator IR.