conv_transpose3d_op

→TensorImplPtr

int conv_transpose3d_op(const int & x, const int & W, const int & b, int stride_d, int stride_h, int stride_w, int pad_d, int pad_h, int pad_w, int opad_d, int opad_h, int opad_w)

source edit

Python wrappers

pylucid.nn.functional.conv_transpose3dfunction

Three-dimensional transposed convolution.

The 3-D extension used in volumetric decoders, video generators and medical-image synthesis. All three spatial axes are upsampled simultaneously.

Parameters

xTensorImplPtr

Input of shape (B, C_in, D, H, W).

WTensorImplPtr

Filters of shape (C_in, C_out, KD, KH, KW).

bTensorImplPtr

Bias of shape (C_out,) or empty.

stride_d, stride_h, stride_wint= None

Per-axis upsampling factor. Default: 1.

pad_d, pad_h, pad_wint= None

Per-axis padding subtracted from each output extent. Default: 0.

opad_d, opad_h, opad_wint= None

Per-axis one-sided output padding. Default: 0.

Returns

TensorImplPtr

Output of shape (B, C_out, D_out, H_out, W_out) with each output extent following $X_\text{out} = (X - 1) s_x - 2 p_x + K_X + p^\text{out}_x$ for $X \in \{D, H, W\}$ .