Abstract
The task of displacement estimation for frames of a video sequence is considered. A new convolutional neural network architecture for the optical flow problem is proposed. The method is based on learning the regularization operator for a fast optimization method. The proposed method has low computational complexity and memory footprint at test time. The neural network architecture is based on unrolling iterations of a fast primal-dual method as layers of a convolutional neural network. Iterations of the optimization method are represented as convolutions with filters that are trained on ground truth data by backpropagation. A real-time implementation using graphics processing units is proposed. Experimental results demonstrate an improved quality of the optical flow field as compared to the optimization method based on a fixed regularization operator.
Similar content being viewed by others
References
Horn, B.K.P. and Schunck, B.G., Determining optical flow, Artif. Intell., 1981, vol. 17, nos. 1–3, pp. 185–203.
Sun, D., Roth, S., and Black, M.J., Secrets of optical flow estimation and their principles, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2010, pp. 2432–2439
Chen, Q. and Koltun, V., Full flow: Optical flow estimation by global optimization over regular grids, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2016, pp. 4706–4714
Menze, M., Heipke, C., and Geiger, A., Discrete optimization for optical flow, Proc. German Conf. Pattern Recognition, 2015, pp. 16–28
Güney, F., Geiger, A., Deep discrete flow, Proc. Asian Conf. Computer Vision, 2016, pp. 207–224
Weinzaepfel, Ph., Revaud, J., Harchaoui, Z., et al., DeepFlow: Large displacement optical flow with deep matching, Proc. IEEE Int. Conf. Computer Vision, 2013, pp. 1385–1392
Hu, Y., Li, Y., and Song, R., Robust interpolation of correspondences for large displacement optical flow, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2017.
Zweig, Sh. and Wolf, L., InterpoNet: A brain inspired neural network for optical flow dense interpolation, Preprint no. 1611.09803, 2016.
Revaud, J., Weinzaepfel, Ph., Harchaoui, Z., et al., Epicflow: Edge-preserving interpolation of correspondences for optical flow, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2015, pp. 1164–1172
Min Ba., Wenjie Luo, Kaustav Kundu, et al., Exploiting semantic information and deep matching for optical flow, Proc. European Conf. Computer Vision, 2016, pp. 154–170
Sevilla-Lara, L., Sun, D., Jampani, V., et al., Optical flow with semantic segmentation and localized layers, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2016, pp. 3889–3898
Hur, J. and Roth, S., Joint optical flow and temporally consistent semantic segmentation, Proc. European Conf. Computer Vision, 2016, pp. 163–177
Dosovitskiy, A., Fischer, Ph., Ilg, E., et al., Flownet: Learning optical flow with convolutional networks, Proc. IEEE Int. Conf. Computer Vision, 2015, pp. 2758–2766
Ilg, E., Mayer, N., Saikia, T., et al., Flownet 2.0: Evolution of optical flow estimation with deep networks, Preprint no. 1612.01925, 2016.
Xu, J., Ranftl, R., and Koltun, V., Accurate optical flow via direct cost volume processing, Preprint no. 1704. 07325, 2017.
Rudin, L.I., Osher, S., and Fatemi, E., Nonlinear total variation based noise removal algorithms, Phys. D (Amsterdam, Neth.), 1992, vol. 60, nos. 1–4, pp. 259–268.
Zach, Ch., Pock, Th., and Bischof, H., A duality based approach for realtime TV-L1 optical flow, Pattern Recognit., 2007, pp. 214–223
Wedel, A., Pock, Th., Zach, Ch., et al., An improved algorithm for TV-L1 optical flow, Statistical and Geometrical Approaches to Visual Motion Analysis, Springer, 2009, pp. 23–45
Bredies, K., Kunisch, K., and Pock, Th., Total generalized variation, SIAM J. Image Sci., 2010, vol. 3, no. 3, pp. 492–526.
Knoll, F., Bredies, K., Pock, Th., et al., Second order total generalized variation (TGV) for MRI, Magn. Reson. Med., 2011, vol. 65, no. 2, pp. 480–491.
Ranftl, R., Gehrig, S., Pock, Th., et al., Pushing the limits of stereo using variational stereo estimation, Proc IEEE Intelligent Vehicles Symp. (IV), 2012, pp. 401–407
Werlberger, M., Convex approaches for high performance video processing, 2012.
Werlberger, M., Pock, Th., and Bischof, H., Motion estimation with non-local total variation regularization, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2010, pp. 2464–2471
Ranftl, R., Bredies, K., and Pock, Th., Non-local total generalized variation for optical flow estimation, Proc. European Conf. Computer Vision, 2014, pp. 439–454
Steinbrucker, F., Pock, Th., and Cremers, D., Large displacement optical flow computation without warping, Proc. 12th IEEE Int. Conf. Computer Vision, 2009, pp. 1609–1614
Schmidt, U. and Roth, S., Shrinkage fields for effective image restoration, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2014, pp. 2774–2781
Lefkimmiatis Stamatios, Non-local color image denoising with convolutional neural networks, Preprint no. 1611.06757, 2016.
Roth, S. and Black, M.J., Fields of experts: A framework for learning image priors, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2005, vol. 2, pp. 860–867.
Wang, Sh., Fidler, S., and Urtasun, R., Proximal deep structured models, Advances in Neural Information Processing Systems, 2016, pp. 865–873
Jaderberg, M., Simonyan, K., Zisserman, A., et al., Spatial transformer networks, Advances in Neural Information Processing Systems, 2015, pp. 2017–2025
Butler, D.J., Wulff, J., Stanley, G.B., et al., A naturalistic open source movie for optical flow evaluation, Proc. European Conf. Computer Vision, 2012. pp. 611–625.
Kingma, D. and Ba, J., Adam: A method for stochastic optimization, Preprint no. 1412.6980, 2014.
Brox, Th. and Malik, J., Large displacement optical flow: Descriptor matching in variational motion estimation, IEEE Trans. Pattern Anal. Mach. Intell., 2011, vol. 33, no. 3, pp. 500–513.
Sundaram, N., Brox, Th., and Keutzer, K., Dense point trajectories by GPU-accelerated large displacement optical flow, Proc. European Conf. Computer Vision, 2010, pp. 438–451
Revaud, J., Weinzaepfel, Ph., Harchaoui, Z., et al., Deepmatching: Hierarchical deformable dense matching, Int. J. Comput. Vision, 2016, vol. 120, no. 3, pp. 300–323.
Meister, S., Hur, J., and Roth, S., UnFlow: Unsupervised learning of optical flow with a bidirectional census loss, Preprint no. 1711.07837, 2017.
Bao Linchao, Yang Qingxion., and Jin Hailin, Fast edge-preserving patchmatch for large displacement optical flow, Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2014, pp. 3534–3541
Brox, Th., Bruhn, A., Papenberg, N., et al., High accuracy optical flow estimation based on a theory for warping, Comput. Vision ECCV, 2004, pp. 25–36
Linchao Bao, Hailin Jin, Byungmoon Kim, et al., A comparison of TV-L1 optical flow solvers on GPU, Proc. GPU Technol. Conf. (GTC) Posters, 2014, pp. 42–54
Author information
Authors and Affiliations
Corresponding author
Additional information
Original Russian Text © A.I. Kuzmin, 2018, published in Programmirovanie, 2018, Vol. 44, No. 3.
Rights and permissions
About this article
Cite this article
Kuzmin, A.I. Learning the Regularization Operator for the Optical Flow Problem. Program Comput Soft 44, 139–147 (2018). https://doi.org/10.1134/S0361768818030040
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1134/S0361768818030040