Deconvolution via Convolution Segments on Standard CNN Hardware
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional hardware lacks native support for deconvolution operations, leading to inefficient performance in applications that rely on deconvolutions, such as machine learning and computer vision, due to the non-trivial nature of deconvolution operations.
Innovation Solution
Implement deconvolution using a set of convolutions, where multiple convolution operations are performed to generate strided segments of the deconvolution result, and the outputs are reassembled to produce the final deconvolution output, utilizing convolution parameters like filter size, stride value, and padding to achieve this.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional hardware is used for deconvolution operations, then device complexity is reduced, but computation speed and resource usage efficiency deteriorate
Solution Approach 1:
The patent makes convolution hardware perform deconvolution operations by reconfiguring the same computational units to execute transposed convolution algorithms, allowing a single hardware architecture to serve multiple functions (convolution and deconvolution) without requiring separate dedicated hardware for each operation type
Solution Approach 2:
The patent changes the operational parameters of convolution hardware by modifying the algorithm implementation (using transposed convolution with adjusted kernel sizes, strides, and padding) to achieve deconvolution functionality, thereby improving computation speed without changing the physical hardware structure
2Productivity
If specialized deconvolution hardware is developed, then computation speed improves, but device complexity and manufacturing cost increase
Solution Approach 1:
The patent eliminates the need for specialized deconvolution hardware by enabling existing convolution hardware to perform deconvolution operations through algorithmic reconfiguration, thereby maintaining computation speed improvements while avoiding increased device complexity and manufacturing costs
3Device complexity
If general-purpose computing hardware is used for deconvolution, then device complexity is reduced, but resource usage efficiency deteriorates
Solution Approach 1:
The patent optimizes resource usage by changing the operational parameters of convolution hardware to efficiently execute deconvolution algorithms, thereby reducing energy consumption and computational overhead compared to using general-purpose hardware
Data Source
AI summary
Disclosed herein are apparatus, method, and computer-readable storage device embodiments for implementing deconvolution via a set of convolutions. For example, an apparatus includes a memory and at least one processor, communicatively coupled with the memory, to perform operations including obtaining a tensor representing image data, generating a deconvolution segment of a plurality of deconvolution segments by performing a convolution to transform the tensor based at least in part on a convolution filter and a stride value, and generating a complete deconvolution for image processing by assembling each deconvolution segment of the plurality of deconvolution segments.