FlowNet
lastmod 2026-11-10created 2019-11-294 edits
FlowNet is a paper and model that proposed estimating optical flow between two frames directly with a convolutional neural network. It presented two architectures: FlowNetSimple, which feeds the two images stacked along the channel dimension, and FlowNetCorr, which extracts features from each image separately and matches them with a correlation layer; it also created the synthetic dataset Flying Chairs for training. It is an early work that solved optical flow with end-to-end learning, and it was followed by the successor model FlowNet2.
FlowNet: Learning Optical Flow with Convolutional Networks.
Documentation
- FlowNet - Learning Optical Flow with Convolutional Networks
- https://arxiv.org/pdf/1504.06852.pdf