Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Learning Optical Flow from a Few Matches

About

State-of-the-art neural network models for optical flow estimation require a dense correlation volume at high resolutions for representing per-pixel displacement. Although the dense correlation volume is informative for accurate estimation, its heavy computation and memory usage hinders the efficient training and deployment of the models. In this paper, we show that the dense correlation volume representation is redundant and accurate flow estimation can be achieved with only a fraction of elements in it. Based on this observation, we propose an alternative displacement representation, named Sparse Correlation Volume, which is constructed directly by computing the k closest matches in one feature map for each feature vector in the other feature map and stored in a sparse data structure. Experiments show that our method can reduce computational cost and memory use significantly, while maintaining high accuracy compared to previous approaches with dense correlation volumes. Code is available at https://github.com/zacjiang/scv .

Shihao Jiang, Yao Lu, Hongdong Li, Richard Hartley• 2021

Related benchmarks

TaskDatasetResultRank
Optical Flow EstimationKITTI 2015 (train)
Fl-epe0.75
446
Optical FlowSintel (train)
AEPE (Clean)0.79
200
Optical Flow EstimationSintel Final (test)
EPE3.6
133
Optical Flow EstimationSintel clean (test)
EPE1.72
120
Optical FlowSintel (test)
AEPE (Final)3.6
120
Optical Flow EstimationKITTI 2015 (test)
Fl-all6.17
108
Optical FlowKITTI (train)
Fl-all0.193
84
Optical FlowMPI Sintel (train)
EPE (Final)2.95
63
Optical Flow EstimationKITTI-15 (test)
Fl-all Error6.17
53
Optical Flow EstimationKITTI (test)
F1-all6.17
20
Showing 10 of 10 rows

Other info

Code

Follow for update