Text-guided object counting aims to count objects specified by textual descriptions in images, but current methods often rely on simple cosine similarity to align visual and textual embeddings, limiting their ability to capture complex relationships. While some studies have explored more advanced methods for constructing similarity maps, they still fall short of effectively leveraging multi-scale and multi-layered information. To address this, we propose AMS-Counter (Adaptive Multi-view Similarity-map Counter), a novel framework that enhances counting accuracy by fusing multi-view features in both spatial and frequency domains. Frequency features effectively represent the structural information necessary for precise image understanding. Building on the Adaptive Multi-view Similarity-map (AMS) to capture multi-layered features, AMS-Counter propose a U-shaped Cross-attention Decoder (UC-Decoder) that effectively integrates information across multiple scales. Evaluations on the FSC-147 benchmark demonstrate that AMS-Counter achieves state-of-the-art performance, with notable improvements in counting accuracy. Code is available: https://github.com/CPSDSC-Lab/AMS-Counter.

In transportation cyber-physical systems, real-time calculations of crowd density can provide traffic managers with important decision support to help optimise traffic flow and prevent congestion and safety accidents. However, crowd density estimation in complex traffic scenarios requires a comprehensive response to the challenges of background noise interference, dense occlusion, and scale variation. To address these issues, we propose a novel cross-level attention multi-scale context-enhanced crowd counting network (CAMCNet). First, we explore the feasibility of the ResNeXt network as a backbone for crowd counting. The residual structure of the ResNeXt model facilitates the generation of hierarchical scale features. Second, the high-level features extracted by this backbone lack detailed information, hindering the acquisition of accurate spatial location information of the crowd. Therefore, we present a high-level feature enhancement module to further refine the perception of multi-scale context and orientation features. Third, we design a cross-level attention feature fusion module to adaptively aggregate enhanced deep features and shallow features to better cope with the challenges of complex background interference and scale variations simultaneously. Finally, to enhance the robustness of network training, we introduce smooth L1 loss into the overall loss function. On four benchmark datasets, extensive experiments show that CAMCNet outperforms the vast majority of the most advanced methods. In particular, CAMCNet achieves state-of-the-art performance on two large-scale mainstream benchmarks (UCF-QNRF and NWPU) in terms of the MAE.

In this paper, we demonstrate that traditional chaotic encryption schemes are vulnerable to the known-plaintext attack (KPA) with deep learning. Considering the decryption process as image restoration based on deep learning, we apply Convolutional Neural Network to perform known-plaintext attack on chaotic cryptosystems. We design a network to learn the operation mechanism of chaotic cryptosystems, and utilize the trained network as the decryption system. To prove the effectiveness, we select three existing chaotic encryption schemes as the attacked targets. The experimental results demonstrate that deep learning can be applied to known-plaintext attack against chaotic cryptosystems successfully. Compared with traditional attack methods for chaotic cryptosystems, the proposed method shows obvious advantages: (1) One neural network may be applied to cryptanalysis of various chaotic cryptosystems, not limited to specific one; (2) the proposed method is significantly convenient and cost-efficient. This paper provides a new idea for the cryptanalysis of chaotic cryptosystems.

For crowd counting, the existing methods usually use an end-to-end approach to directly output the final estimated density map and perform the counts. However, as an intermediate representation, the quality of the estimated density map may significantly affect the counting performance. Therefore, some studies have attempted to optimize the estimated density map with additional attention mechanism. But these methods only focus on the high-density crowd areas and ignore the optimization of local detail areas. Consequently, we propose a more intuitive and understandable Density Map Dynamic Refinement Network (DDRNet) consisting of Counter and Refiner to further refine the local detail information of the estimated density map. Our training contains two stages. Specifically, for the first stage, Counter generates the initial density map through the feature extraction module and the backend, while Refiner, which consists of convolutional layers with different dilated rates, further refines the output of the former to obtain the final estimated density map in the second stage. Also, due to the different views of Counter and Refiner during training, we design a dynamic joint training strategy to improve counting performance. Extensive experiments on three crowd counting datasets (ShanghaiTech, UCF_CC_50, UCF-QNRF) demonstrate the effectiveness of the proposed model and achieve superior counting results.

Vehicle detection is a challenging task in computer vision. Since the differences among different vehicle shapes are small and not easy to recognize, the common object detection models usually do not work well for vehicle recognition. To select positive samples, the existing anchor-based detection methods usually use high-quality anchor boxes which are very close to the shapes of the vehicles. Such methods often make the prediction strongly dependent on high-quality anchor boxes. At present, the ATSS (Adaptive Training Sample Selection) method has also been used to select high-quality anchor boxes. The main purpose of this method is to automatically set the shapes of the anchor boxes according to the training datasets. However, for vehicle detection, the shapes of the vehicles are usually similar while the scales of the vehicles are quite different, which may influence the performance of the ATSS based vehicle detection. To solve the above problems, we propose a vehicle detection model PSLTNet (Positives Selection with Low Threshold Network). To reduce the prediction instability due to relying on high-quality anchor boxes, PSLTNet applies the positive samples selection with low threshold to enhance the prediction ability for different quality anchor boxes. To improve the identification ability of different vehicle types, PSLTNet adopts a pyramid structure and cascades dilated convolutions at each level. In addition, in order to make the bounding box regression loss be independent of scales, PSLTNet uses GIOU (Generalized Intersection Over Union) to supervise the position regression task of candidate bounding boxes. Finally, to reduce the loss of a large number of simple negatives, PSLTNet uses Focal Loss to supervise the classification task of the background candidate bounding boxes. The Experimental results upon the BIT-Vehicle dataset show that the proposed method can obtain better performance than that of the existing methods.

The existing approaches for crowd counting usually estimate a density map with deep convolutional neural network to obtain the crowd counts. Influenced by the background noises, some approaches may result in incorrect pedestrian heads recognition. Therefore, some approaches try to estimate an attention map to mask background noises. However, since the background noises are complex and stochastic, single attention is of incompetence to recognize them. Consequently, we proposed softer, and more reasonable Cascade Residual Attention Network (CRANet), which cascades several effective residual attention modules to mask background noises. Also, due to pixel-level isolation of Euclidean loss, we designed a novel Pyramid Structural Similarity Loss to train our CRANet. The proposed approach was evaluated on three crowd datasets. Experimental results demonstrated that our approach achieves the state-of-the-art.

In this paper, we focus on measuring the similarity of sentence pair. Noting that a single sentence vector may lose fine-grained semantic information which is important for sentence matching, we propose an embedding matrix to calculate a multi-granularity similarity matrix and find the true semantic alignment of two sentences. We also propose a semantic importance calculation and semantic decomposition that are simple but effective. The proposed model does not require any sparse features or external resources such as WordNet. Compared with other state-of-the-art models, we successfully train in a short time and achieve competitive results on similarity measurement and paraphrase identification tasks. Experiments and visual analysis show the good performance and interpretability of the model.

In software development environment, software companies usually ignore the user requirements validation process in requirement gathering phase, which results in large number of modifications being required in the software maintenance phase to fulfill the customer requirements.Identification of accurate requirements from user stories and determining the effectiveness of work deliverable of software industry has always been a challenging task.In this paper, a new measurement approach for forward engineering completeness for software was introduced by using requirements validation framework.The forward engineering completeness for software was measured in two steps.In the first step, software component structure was developed in order to find the functional and nonfunctional requirements rejected by the customers in the requirement validation framework.In the second step, completeness of software from component-based development was determined in which the following parameters, such as functional, non-functional completeness attributes, were considered in the measurement process, and the unadopted attributes of the reuse code were also considered.Quality level for the attributes were assigned based upon the valuation of interior quality of the source code.Therefore, it resulted in the reduction of development time required for the software and the cost required for the software development was also reduced.A case study was incorporated in this research to explain the measurement process of forward engineering completeness.If the forward engineering code is satisfying the quality standards, then the code is in the completeness form.The attributes of code that negates to be used were considered as unadopted attributes.

Abstract The phenomenon of global software development (GSD) has been adopted by a majority of the software development firms to achieve the significant benefits it offers. However, there are many challenges faced by the GSD organizations, which are mainly related to requirements change management (RCM). The key objective of this study is to identify the challenges of RCM process in GSD domain. The systematic literature review (SLR) approach has been used to investigate the challenges of RCM activities, and a total of 30 challenges were identified. We have further classified the identified challenges in the domain of client and vendor GSD organizations, aiming to provide a clear understanding of the RCM process and its challenges in the context of both types of GSD organizations. The identified challenges were also categorized into three core types according to the organization size (small, medium sized, or large), which highlights the significance of each challenge for a specific organizational size. In addition, the criticality of the identified challenges was assessed using the criteria of challenges having a frequency greater than or equal to 50%. According to the findings of this study, a framework is provided that could help GSD organizations address the problems related to RCM in a GSD environment.

The emerging technologies such as the Internet-of-Things, sensors, communication networks, have been or will be introduced to conventional domotics to provide a wide variety of smart home services to facilitate the household appliances or home cares and improve the lifestyles of people. Currently, smart home system are integrated with different features from product line and equipped with various sensors and actuators to meet the requirements of house occupants by specifying their customised user policies. However, the introduction of features and policies may result in undesired behaviours, and this effect is known as feature interactions. In this study, the authors proposed a Semantic Web-based policy interaction detection method with rules to model smart home services and policies with the aids of ontological analysis in the smart home domain, so as to construct a semantic context for inferring the interaction of policies. The authors focus their work on user policies interaction, which are detected by using the Semantic Web rule language in semantic context. The approach is successfully applied to the smart home system and is able to detect 90 interactions among 32 user policies by automated reasoning with tools support as Protégé and Jess.