“Pulsar-VQM” is a fixed functional hardware accelerator IP
Video encoding and delivery optimizations play a critical role in modern video streaming service market. Optimizations such as content-aware encoding and multi-codec usage strategies increase the service’s QoE and reduce the cost of storing and delivering video so that the streaming services stay competitive in the market.
You can optimize encoding by measuring the visual quality of the transcoded output, such as selecting the bitrate-resolution combination with the highest VMAF score, ensuring efficient data usage without waste.
VMAF quality metric, which mimics the human visual system, has been widely used to measure quality. But, VMAF is based on ML(Machine Learning) and has a very high computational complexity, making it computing resource intensive and slow.
Pulsar-VQM is a fixed-function hardware accelerator IP designed to deliver the highest computing efficiency for measuring the VMAF Video Quality Metric. Compared to software-based VMAF on a CPU, Pulsar-VQM offers superior performance and faster results. Additionally, it efficiently accelerates the calculation of video metrics such as PSNR, SSIM, and VMAF-NEG within a single IP.
IP Specification
- AXI compliant
- One AXI4 Lite port with 32-bit address and 32-bit data for the host to communicate with the IP
- Seven AXI master ports with 64-bit address & 512-bit data to access the external memory
- Input video format/bit-depth
- No limitation of the chroma sampling format and bit-depth of input video
- IP block reads only luminance, assuming 8-bit per sample
- Fusion of the statistics in software
- Hardware IP computes MCPD, APM, and VIF, but software on Host CPU does VMAF score
- VMAF metric is flexibly updated with user’s statics
- HLS (High-Level Synthesis) Design
- HLS design flow is used for IP development
- RTL optimized for a specific target technology and platform can be generated with ease of timing closure
- Target Platform
- ASIC
- FPGA cards for on-premise
- FPGA instances for AWS and MS Azure
