Idea
CBAND API detects and quantifies banding artifacts in compressed videos to improve streaming and display quality for media platforms.
Research Paper
Core Innovation
This paper introduces CBAND, a no-reference video quality evaluator that uses deep neural network embeddings to detect banding artifacts without requiring reference videos. It outperforms prior models in accuracy and efficiency. Additionally, CBAND can be used as a differentiable loss function to improve video debanding models during training.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: growing demand for high-quality video streaming and compression optimization.
Potential Customers & Pain Points
- Video Streaming Platforms Needing Better Compression Quality
- Display Manufacturers Seeking Artifact-Free High-Resolution Content
- Video Codec Developers Optimizing Compression Algorithms
Business Model
Offer CBAND as a SaaS API for video quality assessment and licensing for integration into video encoding and streaming platforms.
Competitive Landscape
- Netflix VMAF
- SSIMPLUS
- Video Clarity
Implementation Challenges
- Integration with existing video pipelines
- Generalization across codecs and content types
- Adoption by industry standards bodies
Validation Strategy
- Conduct pilot tests with streaming platforms to measure quality improvements
- Benchmark CBAND against existing quality metrics on diverse video datasets
- Collaborate with codec developers to optimize debanding using CBAND loss
Research Paper Overview
Subjective and Objective Quality Assessment of Banding Artifacts on Compressed Videos
Summary
This paper addresses the persistent issue of banding artifacts in compressed high-definition videos, which degrade perceptual quality on high-resolution displays. It introduces LIVE-YT-Banding, the first open video dataset with 160 AV1-compressed videos and 7,200 subjective quality ratings from 45 subjects. The authors propose CBAND, a no-reference video quality evaluator leveraging deep neural network embeddings to detect and quantify banding artifacts efficiently and accurately, outperforming prior models. CBAND can also serve as a differentiable loss for optimizing video debanding models. Dataset, code, and models are publicly available.