Super Resolution: The scaler of tomorrow, here today!
summary
Claims that ML super-resolution beats bicubic by measurable VMAF margins like +5 to +10 VMAF points, but at the cost of heavy compute and a GPU requirement. Proven pipelines use the FFmpeg SR filter with models like SRCNN, ESPCN and waifu2x, training on Y plane only and merging U/V separately. Demos include a real-time browser port via WebGL video.js plugin that runs on decent GPUs but remains server‑preferred for broad deployment.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 2:41Why Scale
- 4:35Algorithms
- 8:08Super Resolution
- 11:16Super Resolution Demo
- 18:19The Catch
- 20:26The Future
related talks
21mReal-Time Video Super Resolution for Live Streaming with WebGPU
17mYuZu: Neural-Enhanced Volumetric Video Streaming
24mSuper Resolution with CoreML
21mImage Super-Resolution in JS
07mSuper Resolution with CoreML
27mEmbedded Deep Learning Super Resolution on GStreamer using ONNX Inference Runtime
12mMeasuring live video quality with minimal complexity, now available for everyone!
21mWhat to do after per-title encoding