Secrets of optical flow estimation and their principles

By Deqing Sun, Stefan Roth, Michael J. Black. 2010

1,631 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Learned Video Compression (2018, methodology): “We penalize distortions in all intermediate motion-compensated reconstructions using the Charbonnier loss, known to work well for flow-based distortions [42].”
Optical Flow Estimation Using a Spatial Pyramid Network (2016, methodology): “At each level of the pyramid we solve for the flow using a convolutional network and up-sample the flow to the next pyramid level.”
Action bank: A high-level representation of activity in video (2012, methodology): “ranges between 240-600 seconds, and a modern optical flow method [34] takes more than 24 hours on the same machine.”
Flow Fields: Dense Correspondence Fields for Highly Accurate Large Displacement Optical Flow Estimation (2015, methodology): “We refer to publications like [2, 27, 29] for a detailed overview of optical ﬂow methods and the general principles behind it.”
Layered RGBD scene flow estimation (2015, methodology): “(6) by a sigmoid function and jointly optimize all the support functions using conjugate gradient descent [29, 37].”
Motion estimation via hierarchical block matching and graph cut (2015, methodology): “And l1 norm is used because of its robustness compared to l2 norm [9].”
SfM-Net: Learning of Structure and Motion from Video (2017, neutral): “Instead of directly optimizing over unknown motion parameters, such approaches optimize neural network weights that allow motion prediction in the presence of ambiguities in the given pair of frames.”
A Fully-Connected Layered Model of Foreground and Background Flow (2013, methodology): “The remaining time is largely spent on computing the initial flow fields with Classic+NL [25] in MATLAB.”
Constant Time Weighted Median Filtering for Stereo Matching and Beyond (2013, background): “We show its superiority in various applications such as depth upsampling, clip-art JPEG artifact removal, and image stylization.”
Unifying Flow, Stereo and Depth Estimation (2022, background): “UNDERSTANDING the 3D scene structure and motion from a set of 2D images has been a long-standing goal of computer vision [1], [2].”
VoxelMorph: A Learning Framework for Deformable Medical Image Registration (2018, methodology): “Traditional optical flow approaches typically solve an optimization problem similar to (1) using variational methods [45]–[47].”
On Bayesian Adaptive Video Super Resolution (2014, methodology): “Future work will incorporate the recent developments in each subproblem, such as highorder image prior model [29], nonlocal motion prior model [35], feature matching for fast moving objects [7], [33], [40], and advanced inference methods for estimating the spatially variant blur kernel [9], [39].”
Layered segmentation and optical flow estimation over time (2012, methodology): “We start with the single-layered output from “Classic+NL” [26] and cluster the flow field into 10 layers.”
Generalized Boundaries from Multiple Image Interpretations (2012, methodology): “Optical ﬂow is an important cue for detecting occlusions in video; we use Sun et al.’s publicly available code [23].”
Video Segmentation by Tracking Many Figure-Ground Segments (2013, methodology): “The optical flow is computed using the Classic+NL algorithm [30].”