A Spatio-Temporal Descriptor Based on 3D-Gradients

By Alexander Kläser, Marcin Marszalek, Cordelia Schmid. 2008

2,054 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, 3D Vision & Pose

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 3D Vision & Pose

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.”
SlowFast Networks for Video Recognition (2018, neutral): “Actions can be formulated as spatiotemporal objects and captured by oriented filtering in spacetime, as done by HOG3D [31] and cuboids [10].”
Multimodal Multipart Learning for Action Recognition in Depth Videos (2015, background): “IP detection on depth sequences [14]. Four dimensional surface normals were shown to be very powerful representations of body movements over depth signals [3]. This idea was a generalization of HOG3D [15] into four dimensional depth videos. They quantized the 4D normal vectors of depth surfaces by t”
Differential Recurrent Neural Networks for Action Recognition (2015, methodology): “The parameters for HOG3D are the same as the one used in [15].”
HOPC: Histogram of Oriented Principal Components of 3D Pointclouds for Action Recognition (2014, methodology): “HOG3D [10] 85.”
One-shot-learning gesture recognition using HOG-HOF features (2013, methodology): “Another possible feature descriptor that includes spatio-temporal position of features could be HOG3D (Klaser and Marszalek, 2008), which was applied to this specific dataset by Fanello et al. (2013).”
Action recognition with trajectory-pooled deep-convolutional descriptors (2015, methodology): “Meanwhile several local descriptors have been proposed to represent the 3D volumes extracted around these interest points, such as Histogram of Gradient (HOG), Histogram of Optical Flow (HOF) [17], 3D Histogram of Gradient (HOG3D) [13], and Extended SURF (ESURF) [39].”
C3D: Generic Features for Video Analysis (2014, methodology): “SIFT and HOG are also extended into 3 D into SIFT-3 D [41] and HOG3D [25] for action recognition.”
Actions ~ Transformations (2015, background): “Traditionally, most research in action recognition has focused on learning discriminative classiﬁers on hand-designed features such as HOG3D [22] and IDT [51].”
A Closer Look at Spatiotemporal Convolutions for Action Recognition (2017, background): “Some proposed video representations include spatiotemporal interest points (STIPs) [21], SIFT-3D [27], HOG3D [18], Motion Boundary Histogram [5], Cuboids [6], and ActionBank [26].”
Moments in Time Dataset: One Million Videos for Event Understanding (2018, neutral): “[27] designed histogram features for video.”
Human Action Recognition Using Factorized Spatio-Temporal Convolutional Networks (2015, background): “Since video sequences can naturally be viewed as three-dimensional (3D) spatio-temporal signals, many existing methods seek to develop different spatiotemporal features for representing spatially and temporally coupled action patterns [12, 31, 8, 5].”
Modeling Spatial-Temporal Clues in a Hybrid Deep Learning Framework for Video Classification (2015, background): “Klaser et al. proposed HOG3D by extending the idea of integral images for fast descriptor computation [19].”
Towards Weakly-Supervised Action Localization (2016, methodology): “The human tracks are then classified with HOG-3D descriptors [22].”
Temporal Segment Networks for Action Recognition in Videos (2017, methodology): “…points or trajectories, and a histogram descriptor is computed to capture the appearance and motion information, such as Histogram of Gradient and Histogram of Flow (HOG/HOF) [39], Histogram of Motion Boundary (MBH) [38], 3D Histogram of Gradient (HOG3D) [40], Extended SURF (ESURF) [36], and so on.”
Hierarchical Deep Temporal Models for Group Activity Recognition (2016, methodology): “HOG3D [18], spatio-temporal features [5], object interaction feature [19], and social role features [20] are used as unary component representations.”
Action Classification with Locality-Constrained Linear Coding (2014, methodology): “1) Cell Descriptor: In each cell, a 3D histogram of oriented gradients (HOG) feature [19] is computed by evaluating a gradient vector at each pixel in that cell:”
Unsupervised Learning using Sequential Verification for Action Recognition (2016, background): “Typical features like STIP [17], HOG3D [18], HOF [19], MBH [20] etc. capture spatiotemporal features in a temporal grid-like structure.”
Asynchronous Temporal Fields for Action Recognition (2016, methodology): “Understanding activities and actions has an extensive history [32, 59, 22, 17, 23, 2, 26, 56, 29, 21].”