研究目的
To deploy image and video processing pipelines developed frame-oriented on a stream-oriented hardware platform like an FPGA, reducing development time by enabling rapid testing and debugging.
研究成果
The proposed VLIW-based FPGA fabric effectively bridges the gap between frame-based programming and stream-based processing, significantly reducing development time from hours to seconds for testing iterations. It offers scalability and can be further optimized using HLS tools for higher performance, making it suitable for medical imaging applications with real-time requirements.
研究不足
The framework requires manual transformation of frame-based code to line-based implementations, and buffer sizes must be carefully chosen to avoid bottlenecks or resource overuse. It may not handle all types of image processing filters efficiently, and performance is dependent on FPGA resources and clock frequency.
1：Experimental Design and Method Selection:
The approach involves using an FPGA overlay fabric with softcore VLIW processors and a streaming memory framework to map frame-based image processing algorithms to a stream-based hardware platform. OpenCL is used for programming, and the design includes a specialized memory hierarchy for data movement between processing stages.
2：Sample Selection and Data Sources:
The evaluation uses medical imaging data with a resolution of 960x960 pixels, typical of interventional X-ray systems, to test convolution filters.
3：List of Experimental Equipment and Materials:
A Xilinx VC707 evaluation board with a Virtex 7 FPGA, ρ-VEX VLIW processors, and associated software tools like the pocl framework and VEX toolchain.
4：Experimental Procedures and Operational Workflow:
The platform is synthesized on the FPGA, with cores organized into streams. Image data is divided into vertical stripes, processed line by line using convolution kernels, and performance metrics are measured in terms of cycles per line and frames per second.
5：Data Analysis Methods:
Performance is analyzed by calculating throughput based on clock cycles and frequency, comparing scalability with related works like BioThreads.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
