High Accuracy Optical Flow Estimation Based on a Theory for Warping

By Thomas Brox, Andrés Bruhn, Nils Papenberg, Joachim Weickert. 2004

3,067 citations recorded by the source corpus.

Subfields: Computer Graphics and Computer-Aided Design, 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 Graphics and Computer-Aided Design

In Computer Vision and Pattern Recognition

Citations in the corpus

Hypercolumns for object segmentation and fine-grained localization (2014, methodology): “For example, in optical flow, coarse levels of the image pyramid are good for correspondence, but finer levels are needed for accurate measurement, and a multiscale strategy is used to get the best of both worlds [7].”
Generative Visual Manipulation on the Natural Image Manifold (2016, methodology): “Motion+Color flow algorithm: We then estimate the color and geometric changes by generalizing the brightness constancy assumption in traditional optical flow methods [30,31].”
Long-term recurrent convolutional networks for visual recognition and description (2014, methodology): “Flow is computed with [4] and transformed into a “flow image” by centering x and y flow values around 128 and multiplying by a scalar such that flow values fall between 0 and 255.”
Two-Stream Convolutional Networks for Action Recognition in Videos (2014, methodology): “We used a popular method of [2], which formulates the energy based on constancy assumptions for intensity and its gradient, as well as smoothness of the displacement field.”
Sequence to Sequence -- Video to Text (2015, methodology): “We follow the approach in [8, 9] and first extract classical variational optical flow features [2].”
Video (language) modeling: a baseline for generative models of natural videos (2014, background): “Multi-Scale Prediction: Multiscale, coarse-to-ﬁne approaches are classic in motion estimation models (Brox et al., 2004).”
Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting (2015, methodology): “ROVER calculates the optical flow of consecutive radar maps using the algorithm in [5] and performs semi-Lagrangian advection [4] on the flow field, which is assumed to be still, to accomplish the prediction.”
Convolutional Two-Stream Network Fusion for Video Action Recognition (2016, methodology): “During ﬁnetuning, at each training iteration we sample the T = 5 frames from each of the 96 videos in a batch by randomly sampling the starting frame, and then randomly sampling the temporal stride ( τ ) ∈ [1 , 10] (so operating over a total of between 15 and 50 frames).”
Long-Term Temporal Convolutions for Action Recognition (2016, neutral): “To investigate the dependency of action recognition on the quality of motion estimation, we experiment with three types of ﬂow inputs obtained either directly from the video encoding, referred to as MPEG ﬂow [27], or from two optical ﬂow estimators, namely Farneback [28] and Brox [29].”
Hierarchical Surface Prediction for 3D Object Reconstruction (2017, background): “[2], real-time computation of optical flow is feasible [37].”
FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks (2016, methodology): “[6] it was shown to correspond to a numerical fixed point iteration scheme coupled with a continuation method.”
FlowNet: Learning Optical Flow with Convolutional Networks (2015, neutral): “Many improvements have been introduced [29, 5, 34].”
Finding action tubes (2014, methodology): “motion-CNN: We compute the optical flow signal for each frame, according to [4].”
P-CNN: Pose-Based CNN Features for Action Recognition (2015, methodology): “To construct P-CNN features, we first compute optical flow [4] for each consecutive pair of frames.”
SIFT Flow: Dense Correspondence across Scenes and Its Applications (2011, methodology): “Image alignment, a.k.a. image registration or correspondence, is a broad topic in computer vision, computer graphics and medical imaging, covering stereo, motion analysis, video compression, shape registration, and object recognition.”
Spatio-temporal video autoencoder with differentiable memory (2015, methodology): “To enforce edge preservation, we use a loss function similar to (Brox et al., 2004; Eigen & Fergus, 2015).”
Unsupervised CNN for Single View Depth Estimation: Geometry to the Rescue (2016, background): “…regularizer is known to over-smooth the estimated motion, however a vast literature of more sophisticated edge preserving regularizers with robust penalty functions like [2,22] for which gradients can be computed are at our disposal and can be easily used with our setup to get sharper depth maps.”
Optical Flow Estimation Using a Spatial Pyramid Network (2016, methodology): “Sometimes methods adopt a more generic ﬁlter constancy assumption [1, 9].”