0 views collected around this technical thread.
Bilibili’s AI‑driven frame‑interpolation pipeline upgrades low‑frame-rate videos to smooth high‑frame-rate 1080p playback by optimizing optical‑flow models for large motion, texture and text artifacts, pruning for speed, and deploying via the BVT SDK across on‑demand and live streams.
The Xiaohongshu technical team secured four ICCV 2023 papers—including an oral presentation—introducing an open‑vocabulary video instance segmentation benchmark and model, a one‑shot neural‑radiance‑field avatar method, a test‑time personalized 3D pose forecasting framework, and an MPI‑based realistic optical‑flow generation technique, all achieving state‑of‑the‑art performance.
The article details how AI‑based image animation and face‑swapping decompose video motion into zero‑order rigid and first‑order affine components via Taylor expansion, using unsupervised U‑Net keypoint extraction, sparse-to-dense optical flow conversion, and dense motion networks that learn masks for region‑wise rigidity and non‑rigid deformation.
During the World Cup, Youku used a frame‑rate‑up‑conversion algorithm that synthesizes intermediate frames by combining block‑based motion estimation, optical‑flow refinement, and a rule‑based plus deep‑learning fusion, enabling smooth 25→50 fps video and preserving fast‑moving objects such as a football.
This article explains the fundamentals of optical flow, traces its development from early variational methods to modern deep‑learning models like FlowNet, and discusses practical applications such as video object detection, semantic segmentation, and novel view synthesis, highlighting both technical challenges and future research directions.
This article introduces the fundamentals and evolution of optical flow, covering classic algorithms such as Horn‑Schunck and Lucas‑Kanade, modern deep‑learning approaches like FlowNet, and their practical applications in video detection, semantic segmentation, and novel view synthesis.