Overview

Optical flow estimates a motion vector for each image pixel, helping systems track objects, stabilize video, guide robots, and interpolate or compress frames. Methods range from classical local and variational estimation to deep correlation-based networks such as RAFT, global matching transformers, and models that use pretrained vision features; event-camera methods handle asynchronous motion cues. Training commonly uses synthetic labeled data, while self-supervised approaches exploit photometric, geometric, and temporal consistency when labels are scarce. A central challenge is making accurate estimates transfer to real scenes with occlusion, large motion, lighting changes, and image corruption without excessive computation.

Latest papers

CardsList

64 papers9 papers in the last four weeks, up 29% on the four weeks before. 0.1% of all new papers.

  1. x2x^2-Fusion: Cross-Modality and Cross-Dimension Flow Estimation in Event Edge Space

    Mar 17, 2026Ruishan Guo, Ciyu Ruan, Haoyang Wang +3Scene Flow EstimationOptical Flow

  2. Flow3r: Factored Flow Prediction for Scalable Visual Geometry Learning

    Feb 23, 2026Zhongxiao Cong, Qitao Zhao, Minsik Jeon +13D ReconstructionFeed-Forward 3D Reconstruction

  3. Towards Dependable Optical Flow: Benchmarks for Real-World Spatiotemporal Corruptions

    Nov 22, 2024Zhonghua Yi, Hao Shi, Qi Jiang +5Optical FlowCorruption Robustness