Dense Trajectories and Motion Boundary Descriptors for Action Recognition

By Heng Wang, Alexander Kläser, Cordelia Schmid, Chenglin Liu. 2013

1,794 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

1. Luc Van Gool
2. Trevor Darrell
3. Dacheng Tao
4. Bernt Schiele
5. Li Fei-Fei
6. Cordelia Schmid
7. Mohit Bansal
8. Yoshua Bengio

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Dacheng Tao
3. Xiaogang Wang
4. Ming-Hsuan Yang
5. Trevor Darrell
6. Yi Yang
7. Andrew Zisserman
8. Jitendra Malik

Citations in the corpus

• Multiview Transformers for Video Recognition (2022, background): “Early works [34, 37, 73] relied on hand-crafted features to encode motion and appearance information.”
• Long-term recurrent convolutional networks for visual recognition and description (2014, methodology): “[29] uses dense trajectories [44] and SIFT features as well as temporal context reasoning modeled in a CRF.”
• Delving into Egocentric Actions (2015, positive): “We set up a baseline using local descriptors from Dense Trajectories (DT) [36], a successful video representation for action recognition in a surveillance setting.”
• Movie Description (2016, methodology): “Specifically they use dense trajectory features (Wang et al. 2013) extracted for the clips and CNN features extracted at center frames of the clip.”
• A dataset for Movie Description (2015, methodology): “pping from the visual input to the semantic representation (SR), modeling pairwise dependencies in a CRF using visual classifiers as unaries. The unaries are trained using an SVM on dense trajectories [66]. In the second step [56] translates the SR to a sentence using Statistical Machine Translation ”
• UvA-DARE (Digital Academic Repository) Self-Supervised Video Representation Learning With Odd-One-Out Networks (neutral): “Most of the prior work in action recognition is dedicated to hand-crafted features [18] such as dense trajectory features [15, 21, 41, 42].”
• Deep Multimodal Feature Analysis for Action Recognition in RGB+D Videos (2016, background): “State-of-the-art RGB based features [66, 71] include 2D motion patterns and appearance information of objects and scenes.”
• Video Representation Learning Using Discriminative Pooling (2018, methodology): “For a fair comparison, we also report our best result combining with hand-crafted features (IDT-FV) [50] for HMDB-51.”
• RGB-D-based Human Motion Recognition with Deep Learning: A Survey (2017, methodology): “Wang et al. (2017b) took a different approach and adopted scene flow to extract features that fused the RGB and depth from the onset.”
• Scene Flow to Action Map: A New Representation for RGB-D Based Action Recognition with Convolutional Neural Networks (2017, background): “ flow for action recognition: how to organize the scene flow vectors and how to effectively exploit the spatio-temporal dynamics. Two kinds of motion representations can be identified: Lagrangian motion [63, 64, 30, 44, 66, 43, 72] and Eulerian motion [4, 5, 80, 38, 78, 67, 68, 3]. Lagrangian motion fo”
• C3D: Generic Features for Video Analysis (2014, methodology): “We have studied the compactness and efficiency of the new feature and found that it is more compact and faster to compute compared to the current best hand-crafted [51] and other Conv2D image deep features [43].”
• Coherent Multi-sentence Video Description with Variable Level of Detail (2014, methodology): “Following the experimental setup of [19] we use videos and dense trajectory features [24] published by [20]; we use the same test split for visual recognition and video description.”
• ViViT: A Video Vision Transformer (2021, methodology): “Early video research used hand-crafted features to encode appearance and motion information [41, 69].”
• Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet? (2017, critical): “However, since these accuracies are very low even compared with earlier methods [5, 26], our results indicate that it is difficult to train deep 3D CNNs from scratch on UCF-101, HMDB-51, and ActivityNet.”
• P-CNN: Pose-Based CNN Features for Action Recognition (2015, positive): “proves the performance. 4.3. Dense trajectory features Dense Trajectories (DT) [39] are local video descriptors that have recently shown excellent performance in several action recognition benchmarks [25, 40]. The method first densely samples points which are tracked using optical flow [15]. For each ”
• Rolling-Unrolling LSTMs for Action Anticipation from First-Person Video (2020, background): “Later, Wang et al. [20], [21] introduced dense trajectories to encode local motion and object appearance.”