In modern fashion design field, it is a mainstream practice to generate clothing images by combining sketch and text. However, the image quality generated by existing multimodal methods combining sketches and text descriptions is suboptimal, as the clothing in the generated image often lacks contour accuracy and stylistic coherence. In this paper, we present STCGen, an advanced multimodal framework that uses both sketches and text to generate clothing images with improved contours and more consistent style. First, we introduce the sketch prior embedding module, which processes sketches to extract key structural features and ensure the consistency of contours, thereby enhancing image details. Second, we propose a cross space attention mechanism to address the issue of text information loss and ensure stylistic consistency, thereby enhancing overall image coherence. Finally, we propose a network simplification scheme to reduce complexity without compromising the quality of resulting images. Experimental results demonstrate that our method excels in generating high-fidelity clothing images.

Action recognition in dynamic video remains challenging, particularly when distinguishing between visually similar actions. While existing methods often rely on holistic representations, they overlook the fine-grained details that are significant for accurate classification. We propose a novel Fine-grained Teacher-student CLIP (FT-CLIP) that integrates body part analysis with holistic action recognition through a teacher-student architecture, bridging the gap between fine-grained action parsing and overall action understanding. The teacher model processes individual body parts alongside specialized description to generate part-specific features, which are then aggregated and distilled into the student model. Through knowledge distillation with learnable prompts, the student model effectively learns to capture subtle action distinctions while maintaining efficient inference. FT-CLIP achieves a more nuanced understanding of complex actions by progressing from detailed body part analysis to comprehensive action recognition. Experiments on Kinetics-TPS under a fully-supervised setting and on HMDB51 and UCF101 under a zero-shot setting demonstrate the effectiveness of our method.

Text-based CAPTCHA is widely used in fields such as user identity verification during human-computer interaction in real scenarios. With the development of artificial intelligence, several technologies that automatically bypass CAPTCHAs have emerged, weakening the robustness of CAPTCHAs. In-depth study of adversarial sample technology is needed to further reduce the accuracy of automatic verification code recognition of deep learning models while retaining correct human recognition. We propose a novel method based on gradient adaptive truncation to generate adversarial text-based CAPTCHAs more efficiently. Based on the generated model, our method dynamically adjusts the gradient truncation threshold according to the progress of the perturbation attack method, thereby improving the performance of the sample generation model. On the authoritative dataset, our method is compared with the existing state-of-the-art methods. The results show that our AT-I-FGSM can more effectively reduce the accuracy of automatic recognition models to identify CAPTCHAs and improve the security of CAPTCHAs. At the same time, our method consumes less time in generating CAPTCHAs.

Document images such as statistical reports and scientific journals are widely used in information technology. Accurate detection of table areas in document images is an essential prerequisite for tasks such as information extraction. However, because of the diversity in the shapes and sizes of tables, existing table detection methods adapted from general object detection algorithms, have not yet achieved satisfactory results. Incorrect detection results might lead to the loss of critical information. Therefore, we propose a novel end-to-end trainable deep network combined with a self-supervised pretraining transformer for feature extraction to minimize incorrect detections. To better deal with table areas of different shapes and sizes, we added a dual-branch context content attention module (DCCAM) to high-dimensional features to extract context content information, thereby enhancing the network's ability to learn shape features. For feature fusion at different scales, we replaced the original 3×3 convolution with a multilayer residual module, which contains enhanced gradient flow information to improve the feature representation and extraction capability. We evaluated our method on public document datasets and compared it with previous methods, which achieved state-of-the-art results in terms of evaluation metrics such as recall and F1-score. https://github.com/YongZ-Lee/TD-DCCAM

Learning to capture human motion is essential to 3D human pose and shape estimation from monocular video, which is widely used in many 3D applications. However, the existing methods mainly rely on recurrent or convolutional operation to model such temporal information, which limits the ability to capture non-local contextual relations of human motion and ignores human joint hierarchies. To address this problem, we propose a Global Temporal-Attentive and Joints-Attention network (GTAJA-Net). This method introduces a Global Attention Feature Integration (GAFI) module and a Motion Tree Fusion Decoder (MTFD) module on the basis of a temporally consistent mesh recovery system (TCMR). A GAFI consisting of a collection of temporal features obtains final temporal features carrying spatial information that enhances temporal correlation and refine the features of the current frame. Meanwhile, MTFD aims at modeling the joint level attention. MTFD considers pose estimation as a top-down hierarchical process similar to SMPL kinematic tree. Though conceptually simple, our GTAJA-Net outperforms the state-of-the-art methods on the 3DPW, MPI-INF-3DHP, and Human3.6M benchmark datasets. Our code is available at https://github.com/xiangcece/GTAJA-Net.

In recent years, the growing concern over mobile security and the associated risks posed by mobile malware have prompted an increased focus on utilizing deep learning models for analyzing Android application security. However, the expansion of deep learning model sizes results in an exponential growth of model parameters, demanding significant computing resources for execution. To address this challenge, we propose a lightweight Android malware detection method based on teacher-assistant-student knowledge distillation. Our method enables predicting on local clients, eliminating the need for cloud-base service interactions, and protecting user privacy. We visualize the binary file of the target Android application as an RGB three-channel color image, using ResNeSt50 as the teacher model, and compress it based on knowledge distillation. An assistant model is incorporated to address the issue of insufficient distillation resulting from the significant gap between the teacher and student models. Additionally, we integrate a split-attention mechanism to enhance the ability of the professor model to acquire deep features of malware images. We conduct experiments on Drebin and CICMalDroid 2020 datasets and the results show that the proposed method can ensure that the detection results of student model are more similar to those of the teacher model while reducing model complexity. Our method reduces the number of model parameters by 95% compare to the teacher model while maintaining accuracy. And the accuracy is improved by 0.63% compare to the traditional distillation method.

Abstract Fashion image retrieval is an important branch of image retrieval technology. With the rapid development of online shopping, fashion image retrieval technology has made a breakthrough from text‐based to content‐based. But there is still not a proper deep learning method used for fashion image retrieval. This article proposes a fashion image retrieval framework based on dilated convolutional residual network which consists of two major parts, image feature extraction and feature distance measurement. For image feature extraction, we first extract the shallow features of the input image by a multi‐scale convolutional network, and then develop a novel dilated convolutional residual network to obtain the deep features of the image. Finally, the extracted features are transformed into high‐dimensional features vector by a binary retrieval vector module. For feature distance measurement, we first use PCA to reduce the dimension of the extracted high‐dimensional vectors. Then we propose a mixed distance measurement algorithm combined with cosine distance and Mahalanobis distance to calculate the spatial distance of the feature vectors for similarity ranking, which solves the problems of poor robustness in complex background fashion image retrieval and the inefficiency calculation of Mahalanobis distance. The experimental results show the superiority of our fashion image retrieval framework over existing state‐of‐the‐art methods.

Image-based virtual try-on aims to transfer a clothes onto a person while preserving both person's and cloth's attributes. However, the existing methods to realize this task require a target clothes, which cannot be obtained in most cases. To address this issue, we propose a novel user-friendly virtual try-on network (UF-VTON), which only requires a person image and an image of another person wearing a target clothes to generate a result of the person wearing the target clothes. Specifically, we adopt a knowledge distillation scheme to construct a new triple dataset for supervised learning, propose a new three-step pipeline (coarse synthesis, clothing alignment, and refinement synthesis) for try-on task, and utilize an end-to-end training strategy to further refine the results. In particular, we design a new synthesis network that includes both CNN blocks and swin-transformer blocks to capture global and local information and generate highly-realistic try-on images. Qualitative and quantitative experiments show that our method achieves the state-of-the-art virtual try-on performance.

During the process of pushing patch packets for Android application hotfix, the attacker can hijack and tamper with the dex file due to the lack of adding a digital signature, which leads to code injection with serious consequences. To address the above problems, an dynamic vulnerability detection system based on mitmproxy is primary proposed, which first utilizes mitmproxy to capture all the packets interacted between the client and the server while locating the dex file, then injects the test code into the dex and pushes it to the client for execution using a man-in-the-middle attack, and finally verifies through the log output by the application whether there is a code injection vulnerability. For 1000 applications in the application market, our system successfully detects 34 new unknown applications with dex injection, and the experimental results show that the system is effective in detecting real-world applications with vulnerabilities caused by hotfix.

During the process of pushing patch packets for Android application hotfix, the attacker can hijack and tamper with the dex file due to the lack of adding a digital signature, which leads to code injection with serious consequences. To address the above problems, an dynamic vulnerability detection system based on mitmproxy is primary proposed, which first utilizes mitmproxy to capture all the packets interacted between the client and the server while locating the dex file, then injects the test code into the dex and pushes it to the client for execution using a man-in-the-middle attack, and finally verifies through the log output by the application whether there is a code injection vulnerability. For 1000 applications in the application market, our system successfully detects 34 new unknown applications with dex injection, and the experimental results show that the system is effective in detecting real-world applications with vulnerabilities caused by hotfix.

The recognition of subway driver behavior is an important way for early warning of public safety. The current models of behavior recognition focus on action recognition of target objects in large-scene, which are difficult to apply for the subway driver behavior recognition directly because of space-time constraints. RepC3D model is proposed for recognizing subway driver behaviors in the paper. The model fuse the features of C3D model and RepVGG model. Firstly we preprocess the dataset by cutting the subway driver operation video into short videos, then the preprocessed dataset is adopted as the input of RepC3D model and is downsampled with the multiscale convolution layers of the main network VGG, which is used to extract the effective features of the driver's action behavior. Next, as the feature tranning network,RepC3D model identify and classfy the behaviors of the subway driver from the videos. The experimental result shows that the RepC3D model is btteetter than the C3D model and RepVGG model in terms of recognition accuracy, false detection rate, and missed detection rate, the recognition efficiency is also improved. The dataset is available at https://github.com/wtazyy/Datasets.git.

Image-based virtual try-on systems with the goal of transferring a target clothing item onto the corresponding region of a person have received great attention recently. However, it is still a challenge for the existing methods to generate photo-realistic try-on images while preserving non-target details(Fig. 1). To resolve this issue, we present a novel virtual try-on network, DP-VTON. First, a clothing warping module combines pixel transformation with feature transformation to transform the target clothing. Second, a semantic segmentation prediction module predicts a semantic segmentation map of the person wearing the target clothing. Third, an arm generation module generates arms of the reference image that will be changed after try-on. Finally, the warped clothing, semantic segmentation map, arms image and other non-target details (e.g. face, hair, bottom clothes) are fused together for try-on image synthesis. Extensive experiments demonstrate our system achieves the state-of-the-art virtual try-on performance both qualitatively and quantitatively.1

As one of the specific vision tasks, person re-identification has become a prevalent research topic in the field of multimedia and computer vision. However, existing feature extraction methods, originating from the quality of the bounding boxes which could cause the inhomogeneity and incoherence of person representation for cluttered backgrounds, are difficult to adapt the challenges of the harsh real-world scenarios. This study develops a Triplet person Appearances Parsing Framework (TAPF) which eliminates the surrounding interference factors of bounding boxes for person re-identification. The framework consists of a triplet person parsing network and an integration mechanism for person local and global appearance information. Concretely, the triplet parsing network includes a channel parsing module, a position parsing module and a color parsing module, which are used to extract the person channel parsing descriptor, regional descriptor and color perception descriptor, respectively. Then, a local and global flatten gaussian operations are performed to integrate the person appearance parsing descriptors to obtain more discriminative features for the person representation. The experimental results have been conducted to validate our proposed algorithm can achieve a better performance for person re-identification on several public datasets, i.e., VIPeR and Market-1501, respectively.

Smoke recognition has been actively studied in the computer vision domain. Due to large variance of smoke color, texture and shapes, smoke recognition is a challenging task. Traditional smoke recognition methods are based on handcrafted features. In the past few years, some methods which are based on convolutional neural networks have been proposed that achieved great improvement of smoke recognition. However, previous methods cannot capture the internal structure of smoke well. Manifolds can represent the internal characteristics of smoke data in lower dimensions, reduce the redundancy of data representation, and obtain more discriminative capabilities. In this paper, an end-to-end deep convolutional neural network which based on manifold structure was proposed to capture more discriminative features for smoke recognition. Experimental results show that the proposed method achieves promising results on the public smoke recognition dataset. The proposed method can obtain high detection rate, high accuracy rate and low false alarm rate in the same time.

Due to the nonlinear deformation of nonrigid and nonuniform tissues, it is challenging to accurately measure the displacements of feature points distributed on the inner parts, boundaries, and separatrices of tissue layers. To address this challenge, we propose a feature point matching technique called RAPID to measure MR 2D slice deformation of nonuniform and nonrigid biological tissues. We propose to use the covariance of several neighboring point statistics computed around a keypoint, as the keypoint descriptor. Inspired by the kernel methods, we advocate adopting a Riemannian pseudo kernel to map SPD matrices to a high dimensional Hilbert space, where the Euclidean geometry applies. We compare our RAPID with two existing schemes (i.e., SIFT and SURF). Our experimental results show that our RAPID is superior to SIFT and SURF, because the benefits offered by RAPID are two-fold. First, our RAPID increases the number of matched data points. Second, RAPID substantially improves the key-point matching accuracy of SIFT and SURF.

The large population exerts high burdens to Chinese health census works. In this paper, we propose our PDCAS (pulmonary disease census aiding system) based on medical image grid, which aims to utilize the superiorities of grid technology to improve the efficiency of high-incidence and occupational pulmonary disease census. PDCAS integrates the individual medical information distributed in different hospitalspsila information systems into Medical Information Centre of one area. The census records are classified through one risk rate based cross clustering model to direct the medical diagnosis and review. The main processing algorithms of PDCAS are subdivided and encapsulated as detachable Web services with adapted granularity to support the grid workflow composition corresponding to different pulmonary diseases or aiding aims. The prototype of PDCAS proves the possible improvement of grid technology to diseases census and other data intensive medical applications.

Content-based image retrieval (CBIR) on the Web is an active research topic in recent years. However, a common problem for the content-based search is the difficulty for getting a query image, which dramatically limits the popularity of their application. The ubiquity of camera phones has opened up a new avenue for image-based mobile search and it also gives new challenges. In this paper, a domain-specific CBIR system on the Web with mobile camera-phone is proposed, which further extends our previous work on VAST system and focuses on the specific garment images. It makes full use of the camera-phone's ability that captures and views images anywhere. The key problems related to the wireless network and media characteristics, system usability and users expectations are depicted. The system architecture and flowchart are described also. Based on these, a prototype is implemented.

The goal of segmentation is to partition an image into disjoint regions, in a manner consistent with human perception of the content. For large-scale, general image dataset, however, there are the competing requirements, including not making complex prior assumptions about the scene, having fast speed and good segmentation quality. In this paper, a two-stage method for image segmentation is presented that incorporates the main principles of region-based segmentation and cluster-analysis approaches. The first stage extracts many regions by watershed approach, which provides an initial segmentation. The second stage of the algorithm groups together these primitive regions into meaningful objects to produce the final segmentation results by an improved fuzzy c-means technique. The proposed approach gives a good tradeoff between the easy usability, efficiency and segmentation quality. The experimental results demonstrate the effectiveness of the proposed approach.

Ontology is becoming a key for grid platform to support the composition of heterogeneous resources by means of processing resource description and enactment. MedlmGrid (Medical Image Grid) aims to archive, access, and analyze medical data from distributed healthcare information systems to adapt to the development of healthcare information infrastructure. But the heterogeneities of those systems, especially the semantic gulfs, hamper their interoperations in grid environment. In this paper, we propose an OSIS (ontology-based semantic integration scheme) for MedlmGrid, which adopts a hybrid method to build MedlmGrid ontologies and unifies its information exchange model with HL7 (Health Level 7) v3 protocol. The MedlmGrid ontologies share the same vocabulary to simplify the knowledge discovery and semantic transformation within distributed environment. The rule-based ontology mapping components are also designed to support semantic operations of MedlmGrid. We test the performances of our scheme with simulation experiments to evaluate the feasibility of our approach.