Overview

BLUEDOT(https://blue-dot.io) proposes AI based ‘Perceptual Quality Optimizer’ that pre-processes source video before any encoder to improve coding efficiency, helping encoder not to waste bitrate and enhancing perceptual video quality at given bitrate and various resolution.

FEATURES

• Proprietary AI model which is trainable
• Pretrained AI model enhancing details of source video that affect perceptual quality and reducing details that are not important Improving encoding efficiency
• Built-in down scaler
• Independent with any existing video codec standards (x264, x265, VP9, AV1)
• Light weighted IP performing at very high speed
• Available on various processor incl. CPU, GPU, and FPGA
• Deployable on both On-premise and Cloud

BENEFITS

• Total output video file size is reduced at same video quality
• Use of network bandwidth is also reduced, lowering CDN and storage cost
• Users can stream higher video quality contents at given network bandwidth
• Users can flexibly integrate PQ Optimizer to their computing infra

How We Have Tested

We have chosen AMD/Xilinx Alveo U30 and AWS EC2 VT1 FPGA instance to evaluate our ‘Perceptual Quality Optimizer’ and see how better the perceptual quality is improved while reducing data rate required to encode video.

PQ OPTIMIZER

• Evaluation version available for GPU instance (g4dn.2xlarge)
• Integrated in FFmpeg media framework

H.264 ENCODE

• AWS EC2 VT1 Instance (U30) is used
• Encode video to get maximum video quality to the human eye in most situations (as given in Xilinx Video SDK github site)
• Quality Metrics (PSNR, SSIM, VMAF)

PQ Optimizer Saves ~30% of Data Rate

PQ_1_3.mp4

(Comparison Video)

• Comparing original video and PQ optimized video, on Ave. 28.16% of data rate is saved at same VMAF score after encoding at various bitrate with H.264 in

• At same bitrate, PQ optimized video shows better subjective quality comparing with original video