Search

Few-shot human action recognition, a prominent area in computer vision, has garnered increasing attention and broader use in real-life scenarios. Extracting spatio-temporal skeletal information from human movement videos offers interpretable and data-efficient features. However, existing spatio-temporal feature encoders face challenges such as handling sequence boundaries and coping with noise. In order to solve the above problems, this paper proposes a temporal complement method to optimize the Dynamic Time Warping (DTW) algorithm based on the feature representation of the human skeleton sequence. DTW helps to find optimal alignment between sequences by warping them in the time domain. This is quite useful specially in scenarios where training data is limited. However, DTW has the drawback that the optimal alignment path is highly sensitive to errors in the time series distance matrix. Therefore, we apply a Virtual Adversarial Training method to improve the anti-noise capability of the algorithm. Here, We introduce adversarial perturbations in the training phase to the time series distance matrix, thus incentivizing the model to be resilient to such noise. Our method achieves highest accuracy among protonet, DTW and DASTM methods for the 5-way-1-shot setting for the NTU-S (77.7%), and Kinetics (41.2%) datasets. For the 5-way-5-shot setting, our method achieves highest accuracy of 51.8% for Kinetics dataset when compared with the other approaches.

Based on dimensional memory mechanism and adaptive elite population, this paper proposes a satisfactory and efficient artificial bee colony algorithm (DMABC_elite) to solve optimization problems and train artificial neural networks (ANN). DMABC_elite proposes the concept of adaptive elite population that changes dynamically with the search process, and modifies the search equations for employed and onlooker bee phases on this basis. In addition, a dimensional memory mechanism has been introduced that allows multi-dimensional updates, which improves exploitation and speeds up convergence. Next, a new selection strategy and a Lévy flight-based solution-generating method are introduced in the scout bee phase to enhance the global search ability. Finally, the performance of DMABC_elite on two different problem groups is analyzed experimentally. On the one hand, DMABC_elite is evaluated using 22 classical benchmark functions with different dimensions and CEC 2013 test functions. Compared with basic ABC and nine state-of-the-art ABC variants, DMABC_elite achieved better results, ranking first in all 10-, 30- and 100-dimensional tests across 22 classical benchmark functions and 30-dimensional tests across CEC 2013 test functions. On the other hand, DMABC_elite is compared with traditional backpropagation-based algorithms and other ABC variants when training seven different ANNs. The results show that DMABC_elite is efficient and competitive in training ANNs.

In the practical scene, object detection faces a very complicated situation. The occlusion problem always occurs in actual scene, which may affect the accuracy of object detection, especially for the occluded objects. For the deep models, a larger dataset with sufficient occlusion samples will improve the performance of the object detection models. However, the sample with occlusion problem is too hard to obtain. Therefore, a global average pooling(GAP) based adversarial Faster-RCNN is proposed to generate the hard samples and enhance the performance of object detection algorithm. Sufficient hard samples can be generated with the help of this model. Therefore, the object detection model can be trained adequately for the occluded objects. The hard sample generation is carried out in the space of image feature instead of image generation directly. The class-dependent part is obtained by the GAP network, and it is obscured to generate the feature map of hard sample for model reinforcement training. Therefore, the better object detection model can be trained using a conventional dataset. The Faster-RCNN is adopted as the baseline. The Faster-RCNN and GAP have a joint training to improve the performance of the proposed model. The simulation results exhibit the validation of the proposed algorithm.