Text-based person search aims to retrieve specific individuals across camera networks using natural language descriptions. However, current benchmarks often exhibit biases towards common actions like walking or standing, neglecting the critical need for identifying abnormal behaviors in real-world scenarios. To meet such demands, we propose a new task, text-based person anomaly search, locating pedestrians engaged in both routine or anomalous activities via text. To enable the training and evaluation of this new task, we construct a large-scale image-text Pedestrian Anomaly Behavior (PAB) benchmark, featuring a broad spectrum of actions, e.g., running, performing, playing soccer, and the corresponding anomalies, e.g., lying, being hit, and falling of the same identity. The training set of PAB comprises 1,013,605 synthesized image-text pairs of both normalities and anomalies, while the test set includes 1,978 real-world image-text pairs. To validate the potential of PAB, we introduce a cross-modal pose-aware framework, which integrates human pose patterns with identity-based hard negative pair sampling. Extensive experiments on the proposed benchmark show that synthetic training data facilitates the fine-grained behavior retrieval, and the proposed pose-aware method arrives at 84.93% recall@1 accuracy, surpassing other competitive methods. The dataset, model, and code are available at https://github.com/Shuyu-XJTU/CMP.

Online reinforcement learning (RL) methods are often data-inefficient or unreliable, making them difficult to train on real robotic hardware, especially quadruped robots. So learning robotic tasks from pre-collected data is a promising direction. Agile and stable legged locomotion remains an open issue in its general form. Analogous to the rapid progress of supervised learning in recent years, the combination of offline reinforcement learning (ORL) and realistic datasets has the potential to make breakthroughs in this challenging field. To facilitate the ORL research for real-world applications, we benchmark ten ORL algorithms in the realistic quadrupedal locomotion dataset. The dataset is collected by the classical model predictive control (MPC) method, rather than the online RL method commonly utilized by previous ORL benchmarks. Extensive experimental results show that the best-performing ORL algorithms can achieve competitive performance compared with the online RL, and even surpass it in some tasks. However, there is still a gap between the learning-based methods and classical MPC, especially in terms of stability and task response accuracy. Our benchmark can provide a fertile ground for future application-oriented ORL research.

In this paper, we introduce a large Multi-Attribute and Language Search dataset for text-based person retrieval, called MALS, and explore the feasibility of performing pre-training on both attribute recognition and image-text matching tasks in one stone. In particular, MALS contains 1,510,330 image-text pairs, which is about 37.5 × larger than prevailing CUHK-PEDES, and all images are annotated with 27 attributes. Considering the privacy concerns and annotation costs, we leverage the off-the-shelf diffusion models to generate the dataset. To verify the feasibility of learning from the generated data, we develop a new joint Attribute Prompt Learning and Text Matching Learning (APTM) framework, considering the shared knowledge between attribute and text. As the name implies, APTM contains an attribute prompt learning stream and a text matching learning stream. (1) The attribute prompt learning leverages the attribute prompts for image-attribute alignment, which enhances the text matching learning. (2) The text matching learning facilitates the representation learning on fine-grained details, and in turn, boosts the attribute prompt learning. Extensive experiments validate the effectiveness of the pre-training on MALS, achieving state-of-the-art retrieval performance via APTM on three challenging real-world benchmarks. In particular, APTM achieves a consistent improvement of +6.96 %, +7.68%, and +16.95% Recall@1 accuracy on CUHK-PEDES, ICFG-PEDES, and RSTPReid datasets by a clear margin, respectively. The dataset, model, and code are available at https://github.com/Shuyu-XJTU/APTM.

With the rise of the electric vehicle market, wireless charging technology for electric vehicles as a convenient, fast and safe new charging technology has been focused on research in recent years. Compared with traditional wired charging, wireless charging has great advantages in terms of economy, convenience, safety, and adaptability. To improve the performance of inductive power transfer (IPT) systems, it is essential to pursue an optimized design of the magnetic couplers. Although the result of the Traditional finite element method (FEM) is very accurate, it needs a lot of time. Hence, for the first time, this paper proposes a general data-driven coil design approach based on the combination of a neural network and the multi-objective optimization algorithm. The proposed method significantly accelerates the design process and provides highly compliant design results. A design example of an integrated magnetic coupler is provided to validate the superiority of the proposed method in time-saving and accuracy, where the FEM simulation results show that the design deviation is within 10%.

Quadruped robots possess advantages on different terrains over other types of mobile robots by virtue of their flexible choices of foothold points. It is crucial to integrate terrain perception with motion planning to exploit the potential of quadruped robots. We propose a novel hierarchical terrain-aware control (HTC) framework, which leverages deep reinforcement learning (DRL) for the high-level planner and optimal control for the low-level controller. In general, traditional control methods yield better stability by using an optimization algorithm. In addition, DRL is able to offer more adaptive behavior. Our approach makes full use of the advantages of these two methods and possesses better adaptability and stability in challenging natural environments. Furthermore, the global height map of the terrain serves as visual information for the DRL, which determines the desired footholds for the robot’s leg swings and body postures. Optimal control calculates the torque of the joints on the standing legs to maintain body balance. Our method is tested on various terrains both simulated and real environments. The experimental results show that HTC can effectively enhance the adaptability of the quadruped robot by coordinating body posture.

This work aims at automated segmentation of major lesions observed in early stages of cervical cancer which is the second most common cancer among women worldwide. The purpose of segmentation is to automatically determine the location for a biopsy to be taken for diagnosis, a process that is currently done manually. The acetowhite region, a major indicator of abnormality in the cervix image, is first segmented by using a nonconvex optimization approach. Within the acetowhite region, other abnormal features such as the mosaic patterns are then automatically classified from nonmosaic regions by texture analysis. The abnormal features are obtained from skeletonized vascular structures uniquely representing typical vascularity embedded in the normal and abnormal regions extracted by a series of mathematical morphological operations. From the extracted vascular structure a texture feature is identified which clearly distinguishes between the normal and abnormal regions and is used for automated segmentation of the mosaic patterns using fuzzy c-means. Analysis and interpretation of cervix images are important in early detection of cervical lesions. Automated image analysis provides quantitative description of lesions thus less subjective variability in monitoring of chronic lesions so that cervical cancer can be treated effectively at its onset

Medical images usually require higher fidelity than commonly used natural images, especially with respect to detail preservation. Color medical images also require a higher degree of color preservation. In this paper, we present the extension of the novel gray-scale image compression technique, hybrid multi-scale vector quantization (HMVQ) to color image compression. Limitations of common color image compression methods in controlling bit-rate allocation for color channels are discussed, and the ability of HMVQ in overcoming such limitations while improving the color quality is demonstrated.

The challenge in producing a clear retinal image lies in the design of the imaging system that can reduce the strong aberrations of the human eye and the photographer's expertise in adjusting the fundus camera when the image is taken. A technique that combines an advanced adaptive optical system and image post-processing algorithm has been shown to be able to significantly improve the image quality up to 2-5 times higher resolution than currently used clinical devices. The optical system has been designed to suppress low-order aberrations and enhance the precision of the wavefront measurement. Image post-processing mainly involves deconvolution by calculating the system transfer function (here the optical transfer function (OTF)) from the data obtained by the Shack-Hartman wavefront sensor.

The significance and need for expert interpretation of cervigrams in the study of human papillomavirus (HPV) are currently being investigated. Results of these preliminary studies suggest development and integration of new optical probes for detection and discrimination of cervical neoplasia using automated image analysis tools to reduce subjective variability and to provide remote areas with effective screening tools for early detection of cervical cancer. However, long-term studies using already available cervical images are needed to validate the potential of automated classification and recognition algorithms in discriminating cervical neoplasia and normal tissue. For the effective dissemination of cervical image data over the Web from a central repository to various study groups, it is essential that the image file size be reduced by advanced color data compression techniques while preserving crucial features of color and spatial details. We present the preliminary results of the effectiveness of a novel, wavelet-based, multi-spectral codec in retaining diagnostic features in encoded cervical images.

Efficient retrieval of digital medical images over the Internet for worldwide users in a platform-independent manner is becoming essential in clinical research and education nowadays. Due to the large size of most X-ray images, traditionally the images are sub-sampled for viewing at a smaller size and with reduced quality. We present a new and efficient way of encoding and decoding large X-ray images for Internet users without any appreciable reduction in image quality and no reduction in size by a high-fidelity vector quantizer in the wavelet domain. As opposed to a 16:1 reduction in size by sub-sampling, this new technique decreases the file size to over 100:1 with insignificant loss and no reduction in size in the decoded image viewed by Web users anywhere. In addition to traditional vector quantization in the wavelet domain, a new scheme for generating the vectors from images is used. In the proposed scheme, a wavelet transform is obtained by a lifting scheme, and the vectors are generated over a multi-scale wavelet domain by relating wavelet coefficients from coarser to finer resolution. Such a multi-resoultion tree for scalar quantization is well-known but it has never been used in vector quantization.

Fourier transform IR microspectroscopy (FTIRM) of bone slices has become a major research tool to analyze the chemical constitution of calcified structures. Qualitative as well quantitative measures of bone components using this technique have been shown to be useful in the characterization of normal bone in both animal and human specimens. Furthermore, the technique can be used to characterize bone diseases like osteoporosis and the effects of different chemical constitutions on bone structures. However, the ever-increasing number of FTIRM studies is yielding an amount of data that is becoming unmanageable in terms of size for the purpose of storage. We address this problem by applying two different lossless coding techniques to the images generated by the studies. Compression ratios as high as 100:1 are obtained by coding the lower bit levels of the image while retaining the information that is used by the specialists.

The performance of compression algorithms is traditionally evaluated by statistical metrics such as mean square error (MSE) and peak signal-to-noise ratio (PSNR). However it is well known that these metrics do not always correlate with visual perception. We have implemented a well known perceptual metric and evaluated the performance of a number of recently developed high fidelity compression algorithms using perceptual as well as statistical metrics. Our preliminary results do not indicate a consistent correlation between the perceptual metric used and subjective ranking.

The generation of multi-resolution codebooks for encoding and decoding large (2K/spl times/1K) color images is, in general, a challenging task, due to the long search processes involved. We have used an adaptive clustering technique for generating multi-resolution codebooks by vector quantization of wavelet-transformed sub-images up to four levels of decomposition in an efficient manner. Using a combination of clustering up to the fourth level and selective retention of sub-images at the first and second levels, images of varying qualities can be reconstructed from a multi-resolution codebook. The histograms of the reconstructed images demonstrate little or no change from the histogram of the original, even up to a compression ratio of 132:1. The peak signal-to-noise ratio and the mean square error are also better than those from other benchmark compression techniques.