TV-L1 Optical Flow Estimation

By Javier Sánchez Pérez, Enric Meinhardt, Gabriele Facciolo. Image Processing On Line, 2013

432 citations recorded by the source corpus.

This article describes an implementation of the optical flow estimation method introduced by Zach, Pock and Bischof in 2007. This method is based on the minimization of a functional containing a data term using the L 1 norm and a regularization term using the total variation of the flow. The main feature of this formulation is that it allows discontinuities in the flow field, while being more robust to noise than the classical approach by Horn and Schunck. The algorithm is an efficient numerical scheme, which solves a relaxed version of the problem by alternate minimization. Source Code A C implementation of this algorithm is provided. The source code and an online demo are accessible at the web page of this article 1 .

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/43009b83d691ff6cc8973e701081176b388c355f

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

AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions (2017, methodology): “We employ two methods: the dense optical flow estimation method TVL-1 [27], and a CNNbased iterative Deep Flow model which is a variant of the FlowNet-ss network [17].”
Clean-Label Backdoor Attacks on Video Recognition Models (2020, methodology): “For our attack, we ﬁrst inject the RGB trigger to videos, then generate optical ﬂow of these poisoned videos using TVL1 algo-rithm [26].”
Real-Time Action Recognition with Enhanced Motion Vector CNNs (2016, methodology): “Our teacher CNN is trained on TV-L1 optical flow [21] that achieves 81.”
Controllable Video Captioning With POS Sequence Guidance Based on Gated Fusion Network (2019, methodology): “For videos, we use TVL1-flow [30] to compute the optical flows in both horizontal and vertical directions for adjacent frames.”
CATER: A diagnostic dataset for Compositional Actions and TEmporal Reasoning (2019, methodology): “For optical ﬂow we use the TVL1 (S´anchez P´erez et al., 2013) implementation.”
Exploiting Spatial-Temporal Modelling and Multi-Modal Fusion for Human Action Recognition (2018, methodology): “To better utilize motion information, we extract optical flows from video sequences not only with the TV L1 algorithm [24] but also with the Farneback algorithm [25].”
Hand-Held Video Deblurring Via Efficient Fourier Aggregation (2015, background): “[44], in particular the implementation given in [45].”
Deep Video Deblurring for Hand-Held Cameras (2017, methodology): “We also use optical flow [37] to align stacks (DBN+FLOW), which is slow to compute and prone to errors (often introducing additional warping artifacts), but allows pixels to be aggregated more easily by removing the spatial variance of corresponding features.”
Spatio-Temporal Filter Adaptive Network for Video Deblurring (2019, methodology): “DVD [36] uses CNN to restore sharp images from neighboring multiple blurry frames, but they use a traditional optical flow method [31] to align these input frames and is computationally expensive.”