English

Learning Correspondence from the Cycle-Consistency of Time

Computer Vision and Pattern Recognition 2019-04-03 v2 Artificial Intelligence Machine Learning

Abstract

We introduce a self-supervised method for learning visual correspondence from unlabeled video. The main idea is to use cycle-consistency in time as free supervisory signal for learning visual representations from scratch. At training time, our model learns a feature map representation to be useful for performing cycle-consistent tracking. At test time, we use the acquired representation to find nearest neighbors across space and time. We demonstrate the generalizability of the representation -- without finetuning -- across a range of visual correspondence tasks, including video object segmentation, keypoint tracking, and optical flow. Our approach outperforms previous self-supervised methods and performs competitively with strongly supervised methods.

Keywords

Cite

@article{arxiv.1903.07593,
  title  = {Learning Correspondence from the Cycle-Consistency of Time},
  author = {Xiaolong Wang and Allan Jabri and Alexei A. Efros},
  journal= {arXiv preprint arXiv:1903.07593},
  year   = {2019}
}

Comments

CVPR 2019 Oral. Project page: http://ajabri.github.io/timecycle

R2 v1 2026-06-23T08:11:52.314Z