Module convolution 

Specialized operations for convolutional neural networks

This module provides efficient implementations of specialized operations that are commonly used in convolutional neural networks, such as im2col/col2im, efficient convolution algorithms, and other related operations.

Functions

col2im

Convert a column matrix back to an input tensor using the col2im algorithm

compute_conv_indices

Compute the indices for batch matrix multiplication in a convolutional layer

conv2d_backward_bias

Apply backward pass of convolution operation for bias gradient

conv2d_backward_input

Apply backward pass of convolution operation for input gradient

conv2d_backward_kernel

Apply backward pass of convolution operation for kernel gradient

conv2d_im2col

Apply convolution operation using im2col and matrix multiplication

conv_transpose2d

Apply 2D transposed convolution (deconvolution) operation

im2col

Extract patches from an input tensor using the im2col algorithm

max_pool2d

Perform max pooling operation on a 4D input tensor

max_pool2d_backward

Perform the backward pass of max pooling operation