S. Ahmadluei; K. Faez; B. Masoumi
Abstract
Deep convolutional neural networks (CNNs) have attained remarkable success in numerous visual recognition tasks. There are two challenges when adopting CNNs in real-world applications: a) Existing CNNs are computationally expensive and memory intensive, impeding their use in edge computing; b) there ...
Read More
Deep convolutional neural networks (CNNs) have attained remarkable success in numerous visual recognition tasks. There are two challenges when adopting CNNs in real-world applications: a) Existing CNNs are computationally expensive and memory intensive, impeding their use in edge computing; b) there is no standard methodology for designing the CNN architecture for the intended problem. Network pruning/compression has emerged as a research direction to address the first challenge, and it has proven to moderate CNN computational load successfully. For the second challenge, various evolutionary algorithms have been proposed thus far. The algorithm proposed in this paper can be viewed as a solution to both challenges. Instead of using constant predefined criteria to evaluate the filters of CNN layers, the proposed algorithm establishes evaluation criteria in online manner during network training based on the combination of each filter’s profit in its layer and the next layer. In addition, the novel method suggested that it inserts new filters into the CNN layers. The proposed algorithm is not simply a pruning strategy but determines the optimal number of filters. Training on multiple CNN architectures allows us to demonstrate the efficacy of our approach empirically. Compared to current pruning algorithms, our algorithm yields a network with a remarkable prune ratio and accuracy. Despite the relatively high computational cost of an epoch in the proposed algorithm in pruning, altogether it achieves the resultant network faster than other algorithms.
V. Fazel Asl; B. Karasfi; B. Masoumi
Abstract
In this article, we consider the problems of abnormal behavior detection in a high-crowded environment. One of the main issues in abnormal behavior detection is the complexity of the structure patterns between the frames. In this paper, social force and optical flow patterns are used to prepare the system ...
Read More
In this article, we consider the problems of abnormal behavior detection in a high-crowded environment. One of the main issues in abnormal behavior detection is the complexity of the structure patterns between the frames. In this paper, social force and optical flow patterns are used to prepare the system for training the complexity of the structural patterns. The cycle GAN system has been used to train behavioral patterns. Two models of normal and abnormal behavioral patterns are used to evaluate the accuracy of the system detection. In the case of abnormal patterns used for training, due to the lack of this type of behavioral pattern, which is another challenge in detecting the abnormal behaviors, the geometric techniques are used to augment the patterns. If the normal behavioral patterns are used for training, there is no need to augment the patterns because the normal patterns are sufficient. Then, by using the cycle generative adversarial nets (cycle GAN), the normal and abnormal behaviors training will be considered separately. This system produces the social force and optical flow pattern for normal and abnormal behaviors on the first and second sides. We use the cycle GAN system both to train behavioral patterns and to assess the accuracy of abnormal behaviors detection. In the testing phase, if normal behavioral patterns are used for training, the cycle GAN system should not be able to reconstruct the abnormal behavioral patterns with high accuracy.
