nz::nodes::calc::Col2ImgNode Class Reference

Reconstructs spatial tensors from column matrices generated by im2col transformation. More...

Inheritance diagram for nz::nodes::calc::Col2ImgNode:

Collaboration diagram for nz::nodes::calc::Col2ImgNode:

Public Member Functions
	Col2ImgNode (Node *input, Tensor::size_type outputHeight, Tensor::size_type outputWidth)
	Constructor for the Col2ImgNode class.
void	forward () override
	Performs the forward propagation operation in the Col2ImgNode.
void	backward () override
	Performs the backward propagation operation in the Col2ImgNode.
Public Member Functions inherited from nz::nodes::Node
virtual void	print (std::ostream &os) const
	Prints the type, data, and gradient of the node.
void	dataInject (Tensor::value_type *data, bool grad=false) const
	Injects data into a relevant tensor object, optionally setting its gradient requirement.
template<typename Iterator >
void	dataInject (Iterator begin, Iterator end, const bool grad=false) const
	Injects data from an iterator range into the output tensor of the InputNode, optionally setting its gradient requirement.
void	dataInject (const std::initializer_list< Tensor::value_type > &data, bool grad=false) const
	Injects data from a std::initializer_list into the output tensor of the Node, optionally setting its gradient requirement.

Detailed Description

Reconstructs spatial tensors from column matrices generated by im2col transformation.

This node performs the inverse operation of Img2ColNode, converting (N, 1, Hout*Wout, Cout) tensors back to (N, Cout, Hout, Wout) spatial format. Essential for gradient propagation in convolution operations and feature map reconstruction.

Core functionality and characteristics:

• Spatial Reconstruction: Transforms column-organized data back to original spatial dimensions.
• Gradient Handling: Reconstructs gradients during backpropagation through reverse im2col.
• Parameter Consistency: Requires matching kernel/stride/padding with original Img2ColNode.
• Memory Optimization: Implements efficient gradient summation for overlapping regions.
• Shape Restoration: Recovers original spatial dimensions through inverse indexing.
• CUDA Support: GPU-accelerated implementation for both forward and backward passes.

Key implementation aspects:

• Forward Pass: Reshapes column matrix to spatial format using output dimensions.
• Backward Pass: Applies im2col transformation to gradients for upstream propagation.
• Overlap Resolution: Accumulates gradients from overlapping input regions during backward pass.
• Shape Validation: Verifies input tensor conforms to (N, 1, H*W, C) structure.

Typical use cases:

• Gradient computation for convolution layers using matrix-based implementations.
• Feature map reconstruction after channel-wise operations.
• Custom layer implementations requiring spatial/tensor format conversion.
• Visualization of intermediate column-organized features.

Critical considerations:

• Parameter Matching: Must use identical kernel/stride/padding to corresponding Img2ColNode.
• Memory Footprint: Gradient tensors may require significant memory during backward pass.
• Input Constraints: Strict 4D input tensor requirement with second dimension = 1.
• Spatial Resolution: Output dimensions must match original pre-im2col calculations.

Warning

• Input tensor shape mismatch will cause runtime errors.
• Incorrect kernel/stride parameters may produce distorted output spatial dimensions.
• Large kernel sizes combined with small strides may create memory pressure during backward pass.

Note

• Output dimensions should satisfy: Hout = (H - K_h + 2*padding)/stride + 1
• Backward pass internally uses Img2ColNode's logic for gradient propagation.
• For dilated convolutions, ensure parameter compatibility with forward operations.

See also

Img2ColNode For the corresponding forward transformation

ConvolutionNode For typical usage context in convolutional networks

Usage Example:

// Previous im2col transformation
Img2ColNode im2col(&input, 3, 3, 1, 1);
im2col.forward(); // Output shape: (32, 1, 256*256, 64)
 
// Reconstruct original spatial format
Col2ImNode col2im(&im2col, 3, 3, 1, 1);
col2im.forward();
 
// Reconstructed shape: (32, 64, 256, 256)
std::cout << "Reconstructed shape: " << col2im.output->shape() << std::endl;
 
// Backward pass through inverse transformation
col2im.backward();

Author

Mgepahmge (https://github.com/Mgepahmge)

Date

2023/10/18

Definition at line 3910 of file Nodes.cuh.

Constructor & Destructor Documentation

Col2ImgNode()

nz::nodes::calc::Col2ImgNode::Col2ImgNode	(	Node *	input,
		Tensor::size_type	outputHeight,
		Tensor::size_type	outputWidth )

Constructor for the Col2ImgNode class.

Parameters

input	A pointer to the input Node (host-to-object). This node provides the necessary input data for the Col2ImgNode.
outputHeight	The height of the output image in size_type (host-to-object).
outputWidth	The width of the output image in size_type (host-to-object).

Returns

None

This constructor initializes a Col2ImgNode object. It sets the output height, width, and number of channels based on the input node's output shape. It then adds the input node to the list of input nodes. A new Tensor object is created for the output, with its shape determined by the batch size from the input node's output, the number of output channels, the specified output height, and width. The output tensor also inherits the gradient requirement from the input node's output. Finally, the node type is set to "Col2Img".

Memory management strategy: The constructor uses std::make_shared to allocate memory for the output tensor. The memory will be automatically managed by the smart pointer, and it will be released when the last reference to the tensor is removed. Exception handling mechanism: There is no explicit exception handling in this constructor. However, if memory allocation for the output tensor fails, a std::bad_alloc exception may be thrown.

Exceptions

std::bad_alloc

If memory allocation for the output tensor fails.

Note

• Ensure that the input pointer is valid and points to a properly initialized Node object.
• The performance of this constructor is mainly determined by the memory allocation for the output tensor, which has a time complexity of O(n), where n is the number of elements in the output tensor.

```cpp
Node* inputNode = new Node();
Tensor::size_type outputHeight = 10;
Tensor::size_type outputWidth = 10;
Col2ImgNode col2ImgNode(inputNode, outputHeight, outputWidth);
```

Author

Mgepahmge(https://github.com/Mgepahmge)

Date

2024/07/15

Definition at line 642 of file Nodes.cu.

Member Function Documentation

backward()

void nz::nodes::calc::Col2ImgNode::backward ( )

overridevirtual

Performs the backward propagation operation in the Col2ImgNode.

Parameters

None

Returns

None

This function conducts the backward propagation step for the Col2ImgNode. It first checks if the output tensor of the input node requires gradient computation. If it does, the function calls iCol2imgBackward, passing the gradient tensor of the input node's output, the gradient tensor of the output, the output height, output width, output channels, and the batch size from the input node's output. The iCol2imgBackward function is responsible for computing the gradients and propagating them back to the input.

Memory management strategy: This function does not allocate or deallocate any memory directly. It operates on the existing gradient tensors of the input and output. Exception handling mechanism: There is no explicit exception handling in this function. If the iCol2imgBackward function encounters an error, it may throw an exception, and the specific type of exception depends on the implementation of iCol2imgBackward.

Exceptions

[Exception

type from iCol2imgBackward] If the iCol2imgBackward function encounters an error during execution.

Note

• Ensure that the gradient tensors of the input node's output and the output tensor of the Col2ImgNode are properly initialized before calling this function.
• The performance of this function depends on the implementation of the iCol2imgBackward function. If the iCol2imgBackward function has a time complexity of O(n), where n is the number of elements in the input or output gradient tensors, then this backward propagation step also has a time complexity of O(n).

```cpp
Col2ImgNode col2ImgNode(...); // Assume Col2ImgNode is properly initialized
col2ImgNode.backward();
```

Author

Mgepahmge(https://github.com/Mgepahmge)

Date

2024/07/15

Implements nz::nodes::Node.

Definition at line 660 of file Nodes.cu.

forward()

void nz::nodes::calc::Col2ImgNode::forward ( )

overridevirtual

Performs the forward propagation operation in the Col2ImgNode.

Parameters

None

Returns

None

This function executes the forward propagation step for the Col2ImgNode. It calls the iCol2img function, passing the output tensor's data, the input node's output tensor's data, the output height, output width, output channels, and the batch size from the input node's output. The iCol2img function is responsible for converting the column - formatted data from the input to image - formatted data in the output.

Memory management strategy: This function does not allocate or deallocate any memory directly. It operates on the existing data pointers of the input and output tensors. Exception handling mechanism: There is no explicit exception handling in this function. If the iCol2img function encounters an error, it may throw an exception, but the nature of the exception depends on the implementation of iCol2img.

Exceptions

[Exception

type from iCol2img] If the iCol2img function encounters an error during execution.

Note

• Ensure that the input node's output tensor and the output tensor of the Col2ImgNode are properly initialized before calling this function.
• The performance of this function depends on the implementation of the iCol2img function. If the iCol2img function has a time complexity of O(n), where n is the number of elements in the input or output tensors, then this forward propagation step also has a time complexity of O(n).

```cpp
Col2ImgNode col2ImgNode(...); // Assume Col2ImgNode is properly initialized
col2ImgNode.forward();
```

Author

Mgepahmge(https://github.com/Mgepahmge)

Date

2024/07/15

Implements nz::nodes::Node.

Definition at line 655 of file Nodes.cu.

---

The documentation for this class was generated from the following files:

• D:/Users/Mgepahmge/Documents/C Program/NeuZephyr/include/NeuZephyr/Nodes.cuh
• D:/Users/Mgepahmge/Documents/C Program/NeuZephyr/src/Nodes.cu