Conv

Description

Convolution.

Input

Three inputs:

X: 4D or 5Dtensor. 4D input: float16 or float; 5D input: float16.

W: weight tensor.

B: (optional) bias, 1D tensor.

Output

One output:

Y: output tensor.

Property

auto_pad: (optional) If X is NCHW, the NOTSET, SAME_UPPER, SAME_LOWER, and VALID attributes are supported. If X is NCDHW, only the NOTSET attribute is supported.

dilations: list of four integers when 4Dtensor is input, and list of five integers when 5Dtensor is input. Specifies the dilation rate for dilation convolution. The value range of the H and W dimensions is [1, 255].

group: int32, specifying the number of blocked connections from input channels to output channels. Both input and output channels must be exactly divided by group. Must be set to 1.

pads: list of four integers when 4Dtensor is input, and list of six integers when 5Dtensor is input. Top, bottom, left, and right padding. The value range is [0, 255].

strides: list of four integers when 4Dtensor is input, and list of five integers when 5Dtensor is input. This attribute specifies the strides of the convolution along the height and width. The value range of the H and W dimensions is [1, 63]. By default, the N and C dimensions are set to 1.

Constraints

Input tensor. The value range of the W dimension is [1, 4096].

Weight tensor. The value range of the H and W dimensions is [1, 255].

If the atc tool does not support the --precision_mode=must_keep_origin_dtype parameter, the input type is float or float64.

ONNX Opset Support

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