Large-scale web video event classification by use of Fisher Vectors

By Chen Sun, Ramkant Nevatia. 2013

85 citations recorded by the source corpus.

Subfields: 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 Vision and Pattern Recognition

Citations in the corpus

Action recognition with trajectory-pooled deep-convolutional descriptors (2015, methodology): “We choose Fisher vector [23] to encode the TDDs of a video clip into high dimensional representation as its effectiveness for action recognition has been verified in previous works [38, 27], and then use a linear SVM as the classifer (C = 100).”
Rethinking Spatiotemporal Feature Learning For Video Understanding (2017, background): “Since then, bigger and more realistic datasets have been proposed [37, 43, 19, 41], and hand-crafted features have also improved significantly [49, 44, 48].”
DISCOVER: Discovering Important Segments for Classification of Video Events and Recounting (2014, methodology): “We used Fisher Vector with both first and second order terms, and followed the suggestions in [21] to apply power normalization and l2-normalization to the feature vectors.”
ACTIVE: Activity Concept Transitions in Video Event Classification (2013, methodology): “A similar approach is used in high level video event classification and shows promising results [19].”
Going deeper into action recognition: A survey (2016, background): “In Jain et al. (2013); Xing et al. (2015); Kantorov and Laptev (2014); Sun and Nevatia (2013) VLAD descriptors obtained from spatiotemporal features are employed for action recognition.”
Semantic Aware Video Transcription Using Random Forest Classifiers (2014, methodology): “To learn action detectors, we first extracted motion compensated dense trajectory features with default parameters [31], and encoded the features with Fisher Vectors [28].”
An end-to-end generative framework for video segmentation and recognition (2015, result): “It was shown that FVs often outperform other methods, a result that was further replicated in a separate study [31].”
Fusing shape and motion matrices for view invariant action recognition using 3D skeletons (2017, methodology): “Compared with BoVW, Fisher Vector has the advantage that it maps the characteristics of low dimensional space to high dimensional space by fisher kernel and significant discriminate can be obtained by even small features [16, 17, 18, 19].”
CUHK & SIAT Submission for THUMOS 15 Action Recognition Challenge (2015, methodology): “We then employ Fisher vector to encode these descriptors of a video clip into high dimensional representation as its effectiveness for action recognition has been verified in previous works [9, 17].”
AdaScan: Adaptive Scan Pooling in Deep Convolutional Neural Networks for Human Action Recognition in Videos (2016, unknown): “[35].”
A Robust and Efficient Video Representation for Action Recognition (2015, background): “Feature trajectories (Matikainen et al, 2009; Messing et al, 2009; Sun et al, 2009; Wang et al, 2013a) have been shown to be a good way for capturing the intrinsic dynamics of video data.”
Action and Event Recognition with Fisher Vectors on a Compact Feature Set (2013, unknown): “[39] also use FVs for dense trajectories, but include four types of descriptors (MBH, HOG, HOF and the shape of the trajectories) as well as use a spatial pyramid and a Gaussian kernel, whereas we only use FVs with MBH descriptors and linear classiﬁers, but use more visual words.”
Towards Good Practices for Action Video Encoding (2014, methodology): “In [23], using FV or VLAD can both improve the event classification accuracy by a large margin.”
Webly-Supervised Video Recognition by Mutually Voting for Relevant Web Images and Web Video Frames (2016, unknown): “improved dense trajectories [41]) into video-level representations with Fisher vectors [37].”