Optimal Video Compression Using Pixel Shift Tracking
summary
Claims an AI pipeline that tracks individual pixels with a PIPs persistent independent particles model and a deep contextual network, nullifying redundant pixels, storing only new pixel blocks plus coordinate arrays while keeping the first frame intact. Prototype compresses at ~15 ms/frame and decompresses at ~50 ms/frame, yielding about 84% size reduction with ~4% measured quality loss; single-frame prediction works better than multi-frame and multipoint tracking is proposed next.
AI-written summary · about
chapters (9)
- 0:00Introduction and Speaker Background
- 0:50Overview of Video Compression
- 2:05Traditional vs. AI-Based Compression Methods
- 4:18Pixel Point Tracking Approach
- 6:28Single Point Trajectory Method
- 9:08Multi-Point Trajectory Method
- 11:18Compression and Decompression Steps
- 17:08Results and Performance
- 22:47Future Approaches and Conclusion
related talks
17mEfficient & Frugal Image Data Processing Pipelines
17mGemino: Practical and Robust Neural Compression for Video Conferencing
18mCoVA: Exploiting Compressed-Domain Analysis to Accelerate Video Analytics
19mRethinking the Video Transcoding Pipeline: How Pre-Filtering Holds the Key to Efficiency and Quality
33mArchitecting Scalable Multi-Modal Video Search
22mAI Enhanced GPU Video Coding: Achieving Joint High Compression Efficiency and Throughput
26mNew video coding techniques under consideration for VP10 - the successor to VP9
15mFast Lossless Depth Image Compression