Hardware Argmax Layer Acceleration for CNN Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer vision and image processing systems face challenges with real-time processing due to the computational and memory-intensive demands of convolutional neural networks (CNNs), particularly when handling large and high-resolution images or high-frame-rate video data, leading to inefficiencies and increased power consumption.
Innovation Solution
A hardware acceleration module is used to generate argmax maps based on feature maps, performing predefined hardware operations such as convolutions and rectified linear unit activations, reducing the need for CPU processing and memory transfer, and allowing for faster and more efficient generation of argmax maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If CPU processing is used for argmax map generation, then flexibility and programmability are maintained, but processing speed decreases and power consumption increases
Solution Approach 1:
The system segments processing tasks by separating argmax map generation (handled by hardware acceleration module) from other CNN operations (handled by CPU). This division allows critical path operations to be accelerated in hardware while maintaining software flexibility for other tasks.
Solution Approach 2:
A memory interface acts as an intermediary between the hardware acceleration module and CPU, enabling efficient data transfer and coordination. The interface manages the exchange of feature maps and argmax maps between hardware and software components.
2Productivity
If hardware acceleration module is used for argmax map generation, then processing speed and efficiency improve, but CPU utilization and memory allocation are reduced
Solution Approach 1:
The hardware acceleration module performs argmax map generation autonomously using dedicated hardware circuits, eliminating the need for CPU intervention in this specific task. The module self-manages the computation process, reducing overall system power consumption despite increased local hardware activity.
3Speed
If more hardware resources are allocated for real-time processing, then processing capability improves, but system cost and complexity increase
Solution Approach 1:
The hardware acceleration module is designed with specialized circuits optimized specifically for argmax map generation operations. Rather than building a completely custom CNN processor, the system adds targeted hardware functionality where it is most needed, maintaining overall system simplicity while achieving real-time performance.
Data Source
AI summary
A hardware acceleration module may generate a channel-wise argmax map using a predefined set of hardware-implemented operations. In some examples, a hardware acceleration module may receive a set of feature maps for different image channels. The hardware acceleration module may execute a sequence of hardware operations, including a portion(s) of hardware for executing a convolution, rectified linear unit (ReLU) activation, and/or layer concatenation, to determine a maximum channel feature value and/or argument maxima (argmax) value for a set of associated locations within the feature maps. An argmax map may be generated based at least in part on the argument maximum for a set of associated locations.