The visual recognition and understanding of human actions remain an active research domain of computer vision, being the scope of various research works over the last two decades. The problem is challenging due to its many interpersonal variations in appearance and motion dynamics between humans, without forgetting the environmental heterogeneity between different video images. This complexity splits the problem into two major categories: action classification, recognising the action being performed in the scene, and spatiotemporal action localisation, concerning recognising multiple localised human actions present in the scene. Previous surveys mainly focus on the evolution of this field, from handcrafted features to deep learning architectures. However, this survey presents an overview of both categories and respective evolution within each one, the guidelines that should be followed and the current benchmarks employed for performance comparison between the state-of-the-art methods.

It is known that the kinematics of the human body skeleton reveals valuable information in action recognition. Recently, modeling skeletons as spatio-temporal graphs with Graph Convolutional Networks (GCNs) has been reported to solidly advance the state-of-the-art performance. However, GCN-based approaches exclusively learn from raw skeleton data, and are expected to extract the inherent structural information on their own. This paper describes REGINA , introducing a novel way to RE asoning G raph convolutional networks IN Human A ction recognition. The rationale is to provide to the GCNs additional knowledge about the skeleton data, obtained by handcrafted features, in order to facilitate the learning process, while guaranteeing that it remains fully trainable in an end-to-end manner. The challenge is to capture complementary information over the dynamics between consecutive frames, which is the key information extracted by state-of-the-art GCN techniques. Moreover, the proposed strategy can be easily integrated in the existing GCN-based methods, which we also regard positively. Our experiments were carried out in well-known action recognition datasets and enabled to conclude that REGINA contributes for solid improvements in performance when incorporated to other GCN-based approaches, without any other adjustment regarding the original method. For reproducibility, the REGINA code and all the experiments carried out will be publicly available at https://github.com/DegardinBruno .

Skeleton-based generative modelling is an important research topic to mitigate the heavy annotation process. In this work, we explore the impact of synthetic data on skeleton-based action recognition alongside its evaluation methods for more precise quality extraction. We propose a novel iterative weakly-supervised learning generative strategy for synthesising high-quality human actions. We combine conditional generative models with Bayesian classifiers to select the highest-quality samples. As an essential factor, we designed a discriminator network that, together with a Bayesian classifier relies on the most realistic instances to augment the amount of data available for the next iteration without requiring standard cumbersome annotation processes. Additionally, as a key contribution to assessing the quality of samples, we propose a novel measure based on human kinematics instead of employing commonly used evaluation methods, which are heavily based on images. The rationale is to capture the intrinsic characteristics of human skeleton dynamics, thereby complementing model comparison and alleviating the need to manually select the best samples. Experiments were carried out over four benchmarks of two well-known datasets (NTU RGB+D and NTU-120 RGB+D), where both our framework and model assessment can notably enhance skeleton-based action recognition and generation models by synthesising high-quality and realistic human actions.

Neural Architecture Search (NAS) methods have been successfully applied to image tasks with excellent results. However, NAS methods are often complex and tend to converge to local minima as soon as generated architectures seem to yield good results. This paper proposes GEA, a novel approach for guided NAS. GEA guides the evolution by exploring the search space by generating and evaluating several architectures in each generation at initialisation stage using a zero-proxy estimator, where only the highest-scoring architecture is trained and kept for the next generation. Subsequently, GEA continuously extracts knowledge about the search space without increased complexity by generating several off-springs from an existing architecture at each generation. More, GEA forces exploitation of the most performant architectures by descendant generation while simultaneously driving exploration through parent mutation and favouring younger architectures to the detriment of older ones. Experimental results demonstrate the effectiveness of the proposed method, and extensive ablation studies evaluate the importance of different parameters. Results show that GEA achieves state-of-the-art results on all data sets of NAS-Bench-101, NAS-Bench-201 and TransNAS-Bench-101 benchmarks.