Discriminative Hierarchical Modeling of Spatio-temporally Composable Human Activities

By Ivan Lillo, Á. Soto, Juan Carlos Niebles. 2014

86 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

In Computer Vision and Pattern Recognition

Citations in the corpus

RGB-D-based action recognition datasets: A survey (2016, neutral): “Composable activities Composable activities dataset [35](http://web.”
Space-Time Representation of People Based on 3D Skeletal Data: A Review (2016, background): “[179] Composable Activities BoW Lowlv Dict X X X Wu et al.”
Skeleton-Based Action Recognition Using Spatio-Temporal LSTM Network with Trust Gates (2017, neutral): “Skeleton-based action recognition has been explored in different aspects in recent years [40], [41], [42], [43], [44], [45], [46], [47], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57].”
PKU-MMD: A Large Scale Benchmark for Continuous Multi-Modal Human Action Understanding (2017, background): “Watch-n-Patch [51] and Compostable Activities [19] are the first datasets focusing on the continues sequences and the inner combination of activities in supervised or unsupervised methods.”
A Hierarchical Pose-Based Approach to Complex Action Understanding Using Dictionaries of Actionlets and Motion Poselets (2016, methodology): “We evaluate our method on four action recognition benchmarks: the MSR-Action3D dataset [17], Concurrent Actions dataset [35], Composable Activities Dataset [18], and sub-JHMDB [11].”
Action Genome: Actions As Compositions of Spatio-Temporal Scene Graphs (2019, background): “However, Computer Vision has done little work in explicitly representing the hierarchical structures of actions [50], even though understanding event partonomies can improve tasks like action recognition.”
A discussion on the validation tests employed to compare human action recognition methods using the MSR Action3D dataset (2014, methodology): “Finally, in addition to the described validation methods that are frequently used in the literature, there are other authors that have not been included in any table because either the validation method is unclear [66,67] or the employed settings are not used by more than one author [68,69,70,71].”
AGQA: A Benchmark for Compositional Spatio-Temporal Reasoning (2021, background): “Compositional reasoning is fundamental to understanding visual events [30, 37] and has been sought after recently by a number of papers [47, 55, 57, 13, 24].”
3D skeleton-based human action classification: A survey (2016, methodology): “The method in [116] divides the body into 4 spatial regions: right arm , right leg , left arm , and left leg .”
Moving Poselets: A Discriminative and Interpretable Skeletal Motion Representation for Action Recognition (2015, methodology): “An important disadvantage of all these methods for building mid-level representations for action classification is that, except for [19, 4, 14], the mid-level codebook/classifier is learned separately from the action classifiers using clustering/mining techniques, which might not be discriminative f”
ActivityNet: A large-scale video benchmark for human activity understanding (2015, background): “Another dimension of complexity is addressed by datasets that focus on composable [21] and concurrent [41] activities, but these are constrained with respect to the scene and environment assumptions.”
Deep Bilinear Learning for RGB-D Action Recognition (2018, background): “The temporal dynamics of each skeleton joint [15, 40, 5, 33] and joint pairs [41, 25, 20, 43, 29] are explored for mining the structure motions depicted in the skeleton sequences.”
RGBD Datasets: Past, Present and Future (2016, background): “Composable activities dataset [66] 14 16 693 X e.”
A Review of Human Activity Recognition Methods (2015, background): “FIGURE 8 | Classification of actions from human poses (Lillo et al., 2014).”
Jointly Learning Heterogeneous Features for RGB-D Activity Recognition (2015, methodology): “For a fair comparison, we followed exactly the same leave-one-subject-out experimental setting as in [19], where each time the activity samples performed by 13 participants were all used to train a model and the rest were used for testing.”