Idea
Neural video codec delivering real-time, cross-platform compression with superior quality for consumer video applications.
Research Paper
Core Innovation
This paper introduces MLVC, a neural video codec that explicitly transmits scale parameters via hyperprior to guarantee entropy coding consistency across hardware platforms. It combines architectural improvements and perceptual training to recover coding efficiency while enabling real-time, cross-platform decoding without bit-exact arithmetic.
Why It Matters
Video applications require efficient compression that works reliably across diverse devices without decoding failures. MLVC addresses this by ensuring hardware-robustness and real-time speed, reducing bandwidth and computational costs while maintaining high video quality. This scalability enables broader adoption in video conferencing, streaming, and other consumer video services.
Market Size (TAM)
$20–50B TAM for video compression codecs; $5–10B SAM from video conferencing, streaming, and mobile device markets. Driven by rising video traffic and demand for cross-device compatibility.
Potential Customers & Pain Points
- Video conferencing providers – Need reliable high-quality compression across devices
- Streaming platforms – Require efficient codecs with low latency
- Mobile device manufacturers – Demand optimized codecs for diverse hardware
- Cloud video services – Seek to reduce bandwidth and compute costs
Business Model
Licensing codec technology to device manufacturers and video platform providers; offering SDKs and APIs for integration; potential SaaS for cloud encoding/decoding services.
Competitive Landscape
- HEVC
- AV1
- DCVC-RT
- VVC
Implementation Challenges
- Integration complexity with existing video infrastructure
- Competition from established codec standards
- Hardware variability and optimization challenges
Validation Strategy
- Benchmark MLVC against HEVC and AV1 on real-world video conferencing datasets
- Demonstrate cross-platform decoding consistency on diverse consumer NPUs
- Pilot deployments with video conferencing and streaming partners
- Collect user feedback on subjective video quality and latency
Research Paper Overview
MLVC: Multi-platform Learned Video Codec for Real-World Deployment
Summary
MLVC is a neural video codec designed for cross-platform compatibility and real-time performance on consumer devices. It ensures consistent decoding across diverse hardware by transmitting scale parameters explicitly, overcoming prior quantization issues. MLVC achieves superior compression efficiency and subjective quality while running at 100 FPS on common NPUs, making it practical for widespread deployment in video conferencing and streaming.