Efficient Feature Extraction, Encoding, and Classification for Action Recognition

By Vadim Kantorov, Ivan Laptev. 2014

253 citations recorded by the source corpus.

Subfields: Biomedical Engineering, 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 Biomedical Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Explainable Artificial Intelligence: Understanding, Visualizing and Interpreting Deep Learning Models (2017, methodology): “The last examples demonstrates the explanation of a Fisher Vector / SVM classifier [16], which was trained for predicting human actions from compressed videos.”
Long-Term Temporal Convolutions for Action Recognition (2016, neutral): “To investigate the dependency of action recognition on the quality of motion estimation, we experiment with three types of flow inputs obtained either directly from the video encoding, referred to as MPEG flow [27], or from two optical flow estimators, namely Farneback [28] and Brox [29].”
Unsupervised Semantic Parsing of Video Collections (2015, methodology): “As features, we use the Fisher vector representation of Dense Trajectory like features (HOG, HOF, and MBH) [22].”
End-to-End Learning of Action Detection from Frame Glimpses in Videos (2015, neutral): “2 shows our model structure.”
Multitype Activity Recognition in Robot-Centric Scenarios (2015, background): “Most of the approaches in the literature tackle the problems of third-person activity recognition [9], [10], [11], [12], two persons interactions [2], [13], [14], [15] and group activities [3], [16], [17].”
Learning in high-dimensional multimedia data: the state of the art (2015, background): “An intuitive way is to reduce the number of input variables before a machine learning algorithm can be successfully applied [6, 13, 38, 89].”
Improved Dense Trajectory with Cross Streams (2016, background): “Extracting hand-crafted features [1, 9, 2] along these trajectories can provide effective local descriptors, and encoding these local descriptors with a Fisher vector (FV) [11] or a vector of locally aggregated descriptors (VLAD) [5] can provide an effective video representation [6].”
ECO: Efficient Convolutional Network for Online Video Understanding (2018, methodology): “Kantorov [26] introduced a fast dense trajectory method.”
Going deeper into action recognition: A survey (2016, methodology): “3 Kliper-Gross et al. (2012) Motion Interchange Patterns 29.2 68.5 General 26.9 Sadanand and Corso (2012) Video Wise 76.4 Group Wise 57.9 Oneata et al. (2013) MBH + SIFT + Sqrt + L2 Normalization 54.”
A discriminative CNN video representation for event detection (2014, unknown): “Another variant of VLAD called VLAD-k, which extends the nearest centers with the k-nearest centers, has shown good performance in action recognition [19, 34].”
Watch-n-patch: Unsupervised understanding of actions and relations (2015, methodology): “Different from these approaches, we use the richer human skeleton and RGB-D features rather than the RGB action features [38, 14].”
MARS: Motion-Augmented RGB Stream for Action Recognition (2019, unknown): “We also evaluate the performance of two computationally efficient flow approaches, MPEGFlow [17] and PWCNet [34] on MiniKinetics.”
Hide-and-Seek: Forcing a Network to be Meticulous for Weakly-Supervised Object and Action Localization (2017, background): “Action localization is a well studied problem [29, 6, 51, 21, 23].”
SCSampler: Sampling Salient Clips From Video for Efficient Action Recognition (2019, background): “The first operates directly on compressed video [23, 52, 56], thus eliminating the need for costly decoding.”
Real-Time Action Recognition with Enhanced Motion Vector CNNs (2016, methodology): “With efﬁcient implementation [12], it takes around 360ms to calculate optical ﬂow (Farneback’s Flow [4]) for one frame in CPU.”
Interpretable human action recognition in compressed domain (2017, methodology): “The action recognition model in [12] makes use of HOF, MBHx and MBHy features.”
Compressed Video Action Recognition (2017, background): “Some prior works have utilized signals from compressed video for detection or recognition, but only as a non-deep feature [15, 36, 38, 46].”