A Duality Based Approach for Realtime TV-L1 Optical Flow

By Christopher Zach, Thomas Pock, Horst Bischof. 2007

1,926 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology

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

In Media Technology

Citations in the corpus

Late Temporal Modeling in 3D CNN Architectures with BERT for Action Recognition (2020, methodology): “The optical flow of the frames are extracted with TV-L1 algorithm [35].”
Contrastive Multiview Coding (2019, methodology): “Therefore, we extract it, it+k and ft from two modalities as three views of a video; for optical flow we use the TV-L1 algorithm [82].”
Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset (2017, methodology): “We computed optical flow with a TV-L1 algorithm [35].”
Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (2017, methodology): “For these experiments, we follow the standard setup as described in [10] and extract optical flow features with the TV-L1 approach [40].”
Beyond short snippets: Deep networks for video classification (2015, methodology): “Optical flow is computed from two adjacent frames sampled at 15fps using the approach of [26].”
Temporal Segment Networks: Towards Good Practices for Deep Action Recognition (2016, methodology): “For the extraction of optical flow and warped optical flow, we choose the TVL1 optical flow algorithm [35] implemented in OpenCV with CUDA.”
Spatiotemporal Residual Networks for Video Action Recognition (2016, methodology): “We precompute optical flow [32] before training and store the flow fields as JPEGs (with displacement vectors > 20 pixels clipped).”
Chained Multi-stream Networks Exploiting Pose, Motion, and Appearance for Action Classification and Detection (2017, methodology): “[52], which is a reliable variational method that runs sufficiently fast.”
Action recognition with trajectory-pooled deep-convolutional descriptors (2015, methodology): “We choose the TVL1 optical flow algorithm [40] and use the OpenCV implementation, due to its balance between accuracy and efficiency.”
Convolutional Two-Stream Network Fusion for Video Action Recognition (2016, methodology): “For the temporal net, we use optical flow [1, 37] stacking with L = 10 frames [22].”
Actions ~ Transformations (2015, methodology): “To compute the optical flow, we apply the TVL1 optical flow algorithm [59] and discretize the values to the range of [0, 255].”
Hierarchical Surface Prediction for 3D Object Reconstruction (2017, background): “[2], real-time computation of optical flow is feasible [37].”
A Closer Look at Spatiotemporal Convolutions for Action Recognition (2017, methodology): “This can be explained by noting that R(2+1)D uses Farneback’s optical ﬂow, while I3D uses a more accurate ﬂow, TV-L1 [39] which is, however, one order of magnitude slower than Farneback’s method.”
Scaling Egocentric Vision: The EPIC-KITCHENS Dataset (2018, methodology): “00 Implementation Details: We train both spatial and temporal streams, the latter on dense optical flow at 30fps extracted using the TV-L1 algorithm [52] between RGB frames using the formulation TV-L1(I2t, I2t+3) to eliminate optical flicker, and released the computed flow as part of the dataset.”
TSM: Temporal Shift Module for Efficient Video Understanding (2018, methodology): “To extract the optical flow information between frames, we followed [48] to use the TVL1 optical flow algorithm [55] implemented in OpenCV with CUDA.”
Moments in Time Dataset: One Million Videos for Event Understanding (2018, methodology): “In the interest of performance, we pre-compute optical flow on consecutive frames using an off-the-shelf implementation of TVL1 optical flow algorithm [46] from the OpenCV toolbox [22].”