VEO analyzes video content and computes optimal encoding parameters using VMAF quality measurement and convex hull analysis. Instead of fixed bitrate ladders, it tailors encoding to the content.
Four optimization methods:
- Per-title: custom bitrate ladder per video via R-D curve analysis
- Per-shot: scene detection + Trellis bit allocation across shots
- Segment-level CRF: closed-loop quality control per temporal segment
- Context-aware: device-specific ladders (mobile/desktop/TV) with appropriate VMAF models
Also includes a browser-based comparison player with per-frame VMAF overlay for visual QA.
Built as a learning project to understand the theory and math behind Netflix's and others per-title/per-shot work, BD-Rate analysis, and content-adaptive streaming.
terranvigil•1h ago
Four optimization methods: - Per-title: custom bitrate ladder per video via R-D curve analysis - Per-shot: scene detection + Trellis bit allocation across shots - Segment-level CRF: closed-loop quality control per temporal segment - Context-aware: device-specific ladders (mobile/desktop/TV) with appropriate VMAF models
Also includes a browser-based comparison player with per-frame VMAF overlay for visual QA.
Built as a learning project to understand the theory and math behind Netflix's and others per-title/per-shot work, BD-Rate analysis, and content-adaptive streaming.