Rong Zheng

Renjie Xu, Saiedeh Razavi, Rong Zheng

Video, as a key driver in the global explosion of digital information, can create tremendous benefits for human society. Governments and enterprises are deploying innumerable cameras for a variety of applications, e.g., law enforcement, emergency management, traffic control, and security surveillance, all facilitated by video analytics (VA). This trend is spurred by the rapid advancement of deep learning (DL), which enables more precise models for object classification, detection, and tracking. Meanwhile, with the proliferation of Internet-connected devices, massive amounts of data are generated daily, overwhelming the cloud. Edge computing, an emerging paradigm that moves workloads and services from the network core to the network edge, has been widely recognized as a promising solution. The resulting new intersection, edge video analytics (EVA), begins to attract widespread attention. Nevertheless, only a few loosely-related surveys exist on this topic. The basic concepts of EVA (e.g., definition, architectures) were not fully elucidated due to the rapid development of this domain. To fill these gaps, we provide a comprehensive survey of the recent efforts on EVA. In this paper, we first review the fundamentals of edge computing, followed by an overview of VA. EVA systems and their enabling techniques are discussed next. In addition, we introduce prevalent frameworks and datasets to aid future researchers in the development of EVA systems. Finally, we discuss existing challenges and foresee future research directions. We believe this survey will help readers comprehend the relationship between VA and edge computing, and spark new ideas on EVA.

Yujiao Hao, Boyu Wang, Rong Zheng

Deep neural network models for IMU sensor-based human activity recognition (HAR) that are trained from controlled, well-curated datasets suffer from poor generalizability in practical deployments. However, data collected from naturalistic settings often contains significant label noise. In this work, we examine two in-the-wild HAR datasets and DivideMix, a state-of-the-art learning with noise labels (LNL) method to understand the extent and impacts of noisy labels in training data. Our empirical analysis reveals that the substantial domain gaps among diverse subjects cause LNL methods to violate a key underlying assumption, namely, neural networks tend to fit simpler (and thus clean) data in early training epochs. Motivated by the insights, we design VALERIAN, an invariant feature learning method for in-the-wild wearable sensor-based HAR. By training a multi-task model with separate task-specific layers for each subject, VALERIAN allows noisy labels to be dealt with individually while benefiting from shared feature representation across subjects. We evaluated VALERIAN on four datasets, two collected in a controlled environment and two in the wild.

Keivan Nalaie, Rong Zheng

Multi-object tracking (MOT) is a vital component of intelligent video analytics applications such as surveillance and autonomous driving. The time and storage complexity required to execute deep learning models for visual object tracking hinder their adoption on embedded devices with limited computing power. In this paper, we aim to accelerate MOT by transferring the knowledge from high-level features of a complex network (teacher) to a lightweight network (student) at both training and inference times. The proposed AttTrack framework has three key components: 1) cross-model feature learning to align intermediate representations from the teacher and student models, 2) interleaving the execution of the two models at inference time, and 3) incorporating the updated predictions from the teacher model as prior knowledge to assist the student model. Experiments on pedestrian tracking tasks are conducted on the MOT17 and MOT15 datasets using two different object detection backbones YOLOv5 and DLA34 show that AttTrack can significantly improve student model tracking performance while sacrificing only minor degradation of tracking speed.

Bo Zhang, Zimeng Zhou, Boyu Jiang et al.

In industrial countries, adults spend a considerable amount of time sedentary each day at work, driving and during activities of daily living. Characterizing the seated upper body human poses using mmWave radars is an important, yet under-studied topic with many applications in human-machine interaction, transportation and road safety. In this work, we devise SUPER, a framework for seated upper body human pose estimation that utilizes dual-mmWave radars in close proximity. A novel masking algorithm is proposed to coherently fuse data from the radars to generate intensity and Doppler point clouds with complementary information for high-motion but small radar cross section areas (e.g., upper extremities) and low-motion but large RCS areas (e.g. torso). A lightweight neural network extracts both global and local features of upper body and output pose parameters for the Skinned Multi-Person Linear (SMPL) model. Extensive leave-one-subject-out experiments on various motion sequences from multiple subjects show that SUPER outperforms a state-of-the-art baseline method by 30 -- 184%. We also demonstrate its utility in a simple downstream task for hand-object interaction.

Peiyan Hu, Haodong Feng, Hongyuan Liu et al.

Predicting the evolution of complex physical systems remains a central problem in science and engineering. Despite rapid progress in scientific Machine Learning (ML) models, a critical bottleneck is the lack of expensive real-world data, resulting in most current models being trained and validated on simulated data. Beyond limiting the development and evaluation of scientific ML, this gap also hinders research into essential tasks such as sim-to-real transfer. We introduce RealPDEBench, the first benchmark for scientific ML that integrates real-world measurements with paired numerical simulations. RealPDEBench consists of five datasets, three tasks, eight metrics, and ten baselines. We first present five real-world measured datasets with paired simulated datasets across different complex physical systems. We further define three tasks, which allow comparisons between real-world and simulated data, and facilitate the development of methods to bridge the two. Moreover, we design eight evaluation metrics, spanning data-oriented and physics-oriented metrics, and finally benchmark ten representative baselines, including state-of-the-art models, pretrained PDE foundation models, and a traditional method. Experiments reveal significant discrepancies between simulated and real-world data, while showing that pretraining with simulated data consistently improves both accuracy and convergence. In this work, we hope to provide insights from real-world data, advancing scientific ML toward bridging the sim-to-real gap and real-world deployment. Our benchmark, datasets, and instructions are available at https://realpdebench.github.io/.

Yunkai Yu, Yingying Wang, Rong Zheng

The Internet of Things (IoT) sensors have been widely employed to capture human locomotions to enable applications such as activity recognition, human pose estimation, and fall detection. Motion capture (MoCap) systems are frequently used to generate ground truth annotations for human poses when training models with data from wearable or ambient sensors, and have been shown to be effective to synthesize data in these modalities. However, the representation of older adults, an increasingly important demographic in healthcare, in existing MoCap locomotion datasets has not been thoroughly examined. This work surveyed 41 publicly available datasets, identifying eight that include older adult motions and four that contain motions performed by younger actors annotated as old style. Older adults represent a small portion of participants overall, and few datasets provide full-body motion data for this group. To assess the fidelity of old-style walking motions, quantitative metrics are introduced, defining high fidelity as the ability to capture age-related differences relative to normative walking. Using gait parameters that are age-sensitive, robust to noise, and resilient to data scarcity, we found that old-style walking motions often exhibit overly controlled patterns and fail to faithfully characterize aging. These findings highlight the need for improved representation of older adults in motion datasets and establish a method to quantitatively evaluate the quality of old-style walking motions.

Longyun Liao, Rong Zheng

2D-to-3D human pose lifting is an ill-posed problem due to depth ambiguity and occlusion. Existing methods relying on spatial and temporal consistency alone are insufficient to resolve these problems especially in the presence of significant occlusions or high dynamic actions. Semantic information, however, offers a complementary signal that can help disambiguate such cases. To this end, we propose LangPose, a framework that leverages action knowledge by aligning motion embeddings with text embeddings of fine-grained action labels. LangPose operates in two stages: pretraining and fine-tuning. In the pretraining stage, the model simultaneously learns to recognize actions and reconstruct 3D poses from masked and noisy 2D poses. During the fine-tuning stage, the model is further refined using real-world 3D human pose estimation datasets without action labels. Additionally, our framework incorporates masked body parts and masked time windows in motion modeling, encouraging the model to leverage semantic information when spatial and temporal consistency is unreliable. Experiments demonstrate the effectiveness of LangPose, achieving SOTA level performance in 3D pose estimation on public datasets, including Human3.6M and MPI-INF-3DHP. Specifically, LangPose achieves an MPJPE of 36.7mm on Human3.6M with detected 2D poses as input and 15.5mm on MPI-INF-3DHP with ground-truth 2D poses as input.

Longyun Liao, Rong Zheng, Andrew Mitchell

In this paper, we present the novel task of estimating the extrinsic parameters of a virtual camera relative to a real camera in exercise videos with a mirror. This task poses a significant challenge in scenarios where the views from the real and mirrored cameras have no overlap or share salient features. To address this issue, prior knowledge of a human body and 2D joint locations are utilized to estimate the camera extrinsic parameters when a person is in front of a mirror. We devise a modified eight-point algorithm to obtain an initial estimation from 2D joint locations. The 2D joint locations are then refined subject to human body constraints. Finally, a RANSAC algorithm is employed to remove outliers by comparing their epipolar distances to a predetermined threshold. MirrorCalib achieves a rotation error of 1.82° and a translation error of 69.51 mm on a collected real-world dataset, which outperforms the state-of-art method.

Keivan Nalaie, Rong Zheng

In this paper, we introduce MVSparse, a novel and efficient framework for cooperative multi-person tracking across multiple synchronized cameras. The MVSparse system is comprised of a carefully orchestrated pipeline, combining edge server-based models with distributed lightweight Reinforcement Learning (RL) agents operating on individual cameras. These RL agents intelligently select informative blocks within each frame based on historical camera data and detection outcomes from neighboring cameras, significantly reducing computational load and communication overhead. The edge server aggregates multiple camera views to perform detection tasks and provides feedback to the individual agents. By projecting inputs from various perspectives onto a common ground plane and applying deep detection models, MVSparse optimally leverages temporal and spatial redundancy in multi-view videos. Notably, our contributions include an empirical analysis of multi-camera pedestrian tracking datasets, the development of a multi-camera, multi-person detection pipeline, and the implementation of MVSparse, yielding impressive results on both open datasets and real-world scenarios. Experimentally, MVSparse accelerates overall inference time by 1.88X and 1.60X compared to a baseline approach while only marginally compromising tracking accuracy by 2.27% and 3.17%, respectively, showcasing its promising potential for efficient multi-camera tracking applications.

Mason Kadem, Rong Zheng

Neural networks are growing more capable on their own, but we do not understand their neural mechanisms. Understanding these mechanisms' decision-making processes, or mechanistic interpretability, enables (1) accountability and control in high-stakes domains, (2) the study of digital brains and the emergence of cognition, and (3) discovery of new knowledge when AI systems outperform humans.

Yujiao Hao, Boyu Wang, Rong Zheng

With the prevalence of wearable devices, inertial measurement unit (IMU) data has been utilized in monitoring and assessment of human mobility such as human activity recognition (HAR). Training deep neural network (DNN) models for these tasks require a large amount of labeled data, which are hard to acquire in uncontrolled environments. To mitigate the data scarcity problem, we design CROMOSim, a cross-modality sensor simulator that simulates high fidelity virtual IMU sensor data from motion capture systems or monocular RGB cameras. It utilizes a skinned multi-person linear model (SMPL) for 3D body pose and shape representations, to enable simulation from arbitrary on-body positions. A DNN model is trained to learn the functional mapping from imperfect trajectory estimations in a 3D SMPL body tri-mesh due to measurement noise, calibration errors, occlusion and other modeling artifacts, to IMU data. We evaluate the fidelity of CROMOSim simulated data and its utility in data augmentation on various HAR datasets. Extensive experiment results show that the proposed model achieves a 6.7% improvement over baseline methods in a HAR task.

Shervin Manzuri Shalmani, Fei Chiang, Rong Zheng

Modern neural networks are powerful predictive models. However, when it comes to recognizing that they may be wrong about their predictions, they perform poorly. For example, for one of the most common activation functions, the ReLU and its variants, even a well-calibrated model can produce incorrect but high confidence predictions. In the related task of action recognition, most current classification methods are based on clip-level classifiers that densely sample a given video for non-overlapping, same-sized clips and aggregate the results using an aggregation function - typically averaging - to achieve video level predictions. While this approach has shown to be effective, it is sub-optimal in recognition accuracy and has a high computational overhead. To mitigate both these issues, we propose the confidence distillation framework to teach a representation of uncertainty of the teacher to the student sampler and divide the task of full video prediction between the student and the teacher models. We conduct extensive experiments on three action recognition datasets and demonstrate that our framework achieves significant improvements in action recognition accuracy (up to 20%) and computational efficiency (more than 40%).

Keivan Nalaie, Renjie Xu, Rong Zheng

In surveillance and search and rescue applications, it is important to perform multi-target tracking (MOT) in real-time on low-end devices. Today's MOT solutions employ deep neural networks, which tend to have high computation complexity. Recognizing the effects of frame sizes on tracking performance, we propose DeepScale, a model agnostic frame size selection approach that operates on top of existing fully convolutional network-based trackers to accelerate tracking throughput. In the training stage, we incorporate detectability scores into a one-shot tracker architecture so that DeepScale can learn representation estimations for different frame sizes in a self-supervised manner. During inference, it can adapt frame sizes according to the complexity of visual contents based on user-controlled parameters. To leverage computation resources on edge servers, we propose two computation partition schemes tailored for MOT, namely, edge server only with adaptive frame-size transmission and edge server-assisted tracking. Extensive experiments and benchmark tests on MOT datasets demonstrate the effectiveness and flexibility of DeepScale. Compared to a state-of-the-art tracker, DeepScale++, a variant of DeepScale achieves 1.57X accelerated with only moderate degradation ~2.3\ in tracking accuracy on the MOT15 dataset in one configuration. We have implemented and evaluated DeepScale++ and the proposed computation partition schemes on a small-scale testbed consisting of an NVIDIA Jetson TX2 board and a GPU server. The experiments reveal non-trivial trade-offs between tracking performance and latency compared to server-only or smart camera-only solutions.

Yujiao Hao, Boyu Wang, Rong Zheng

Wearable sensor-based human activity recognition (HAR) has been a research focus in the field of ubiquitous and mobile computing for years. In recent years, many deep models have been applied to HAR problems. However, deep learning methods typically require a large amount of data for models to generalize well. Significant variances caused by different participants or diverse sensor devices limit the direct application of a pre-trained model to a subject or device that has not been seen before. To address these problems, we present an invariant feature learning framework (IFLF) that extracts common information shared across subjects and devices. IFLF incorporates two learning paradigms: 1) meta-learning to capture robust features across seen domains and adapt to an unseen one with similarity-based data selection; 2) multi-task learning to deal with data shortage and enhance overall performance via knowledge sharing among different subjects. Experiments demonstrated that IFLF is effective in handling both subject and device diversion across popular open datasets and an in-house dataset. It outperforms a baseline model of up to 40% in test accuracy.

Yihao Fang, Shervin Manzuri Shalmani, Rong Zheng

The success of deep neural networks (DNN) in machine perception applications such as image classification and speech recognition comes at the cost of high computation and storage complexity. Inference of uncompressed large scale DNN models can only run in the cloud with extra communication latency back and forth between cloud and end devices, while compressed DNN models achieve real-time inference on end devices at the price of lower predictive accuracy. In order to have the best of both worlds (latency and accuracy), we propose CacheNet, a model caching framework. CacheNet caches low-complexity models on end devices and high-complexity (or full) models on edge or cloud servers. By exploiting temporal locality in streaming data, high cache hit and consequently shorter latency can be achieved with no or only marginal decrease in prediction accuracy. Experiments on CIFAR-10 and FVG have shown CacheNet is 58-217% faster than baseline approaches that run inference tasks on end devices or edge servers alone.

Yongyong Wei, Rong Zheng

Large-scale spatial data such as air quality, thermal conditions and location signatures play a vital role in a variety of applications. Collecting such data manually can be tedious and labour intensive. With the advancement of robotic technologies, it is feasible to automate such tasks using mobile robots with sensing and navigation capabilities. However, due to limited battery lifetime and scarcity of charging stations, it is important to plan paths for the robots that maximize the utility of data collection, also known as the informative path planning (IPP) problem. In this paper, we propose a novel IPP algorithm using reinforcement learning (RL). A constrained exploration and exploitation strategy is designed to address the unique challenges of IPP, and is shown to have fast convergence and better optimality than a classical reinforcement learning approach. Extensive experiments using real-world measurement data demonstrate that the proposed algorithm outperforms state-of-the-art algorithms in most test cases. Interestingly, unlike existing solutions that have to be re-executed when any input parameter changes, our RL-based solution allows a degree of transferability across different problem instances.

Pooya Khandel, Amir Hossein Rassafi, Vahid Pourahmadi et al.

In IoT solutions, it is usually desirable to collect data from a large number of distributed IoT sensors at a central node in the cloud for further processing. One of the main design challenges of such solutions is the high communication overhead between the sensors and the central node (especially for multimedia data). In this paper, we aim to reduce the communication overhead and propose a method that is able to determine which sensors should send their data to the central node and which to drop data. The idea is that some sensors may have data which are correlated with others and some may have data that are not essential for the operation to be performed at the central node. As such decisions are application dependent and may change over time, they should be learned during the operation of the system, for that we propose a method based on Advantage Actor-Critic (A2C) reinforcement learning which gradually learns which sensor's data is cost-effective to be sent to the central node. The proposed approach has been evaluated on a multi-view multi-camera dataset, and we observe a significant reduction in communication overhead with marginal degradation in object classification accuracy.

Chao Cai, Rong Zheng, Jun Luo

With the proliferation of Internet-of-Things devices, acoustic sensing attracts much attention in recent years. It exploits acoustic transceivers such as microphones and speakers beyond their primary functions, namely recording and playing, to enable novel applications and new user experiences. In this paper, we present the first systematic survey of recent advances in active acoustic sensing using commodity hardware with a frequency range below 24~\!kHz. We propose a general framework that categorizes main building blocks of acoustic sensing systems. This framework encompasses three layers, i.e., physical layer, core technique layer, and application layer. The physical layer includes basic hardware components, acoustic platforms as well as the air-borne and structure-borne channel characteristics. The core technique layer encompasses key mechanisms to generate acoustic signals (waveforms) and to extract useful temporal, spatial and spectral information from received signals. The application layer builds upon the functions offered by the core techniques to realize different acoustic sensing applications. We highlight unique challenges due to the limitations of physical devices and acoustic channels and how they are mitigated or overcame by core processing techniques and application-specific solutions. Finally, research opportunities and future directions are discussed to spawn further in-depth investigation on acoustic sensing.

Yongyong Wei, Cristian Frincu, Rong Zheng

Fingerprint-based indoor localization methods are promising due to the high availability of deployed access points and compatibility with commercial-off-the-shelf user devices. However, to train regression models for localization, an extensive site survey is required to collect fingerprint data from the target areas. In this paper, we consider the problem of informative path planning (IPP) to find the optimal walk for site survey subject to a budget constraint. IPP for location fingerprint collection is related to the well-known orienteering problem (OP) but is more challenging due to edge-based non-additive rewards and revisits. Given the NP-hardness of IPP, we propose two heuristic approaches: a Greedy algorithm and a genetic algorithm. We show through experimental data collected from two indoor environments with different characteristics that the two algorithms have low computation complexity, can generally achieve higher utility and lower localization errors compared to the extension of two state-of-the-art approaches to OP.

Yihao Fang, Rong Zheng, Xiaodan Zhu

A novel logographic subword model is proposed to reinterpret logograms as abstract subwords for neural machine translation. Our approach drastically reduces the size of an artificial neural network, while maintaining comparable BLEU scores as those attained with the baseline RNN and CNN seq2seq models. The smaller model size also leads to shorter training and inference time. Experiments demonstrate that in the tasks of English-Chinese/Chinese-English translation, the reduction of those aspects can be from $11\%$ to as high as $77\%$. Compared to previous subword models, abstract subwords can be applied to various logographic languages. Considering most of the logographic languages are ancient and very low resource languages, these advantages are very desirable for archaeological computational linguistic applications such as a resource-limited offline hand-held Demotic-English translator.

Qiang Xu, Rong Zheng, Ezzeldin Tahoun

Mobile crowdsensing allows a large number of mobile devices to measure phenomena of common interests and form a body of knowledge about natural and social environments. In order to get location annotations for indoor mobile crowdsensing, reference tags are usually deployed which are susceptible to tampering and compromises by attackers. In this work, we consider three types of location-related attacks including tag forgery, tag misplacement, and tag removal. Different detection algorithms are proposed to deal with these attacks. First, we introduce location-dependent fingerprints as supplementary information for better location identification. A truth discovery algorithm is then proposed to detect falsified data. Moreover, visiting patterns are utilized for the detection of tag misplacement and removal. Experiments on both crowdsensed and emulated dataset show that the proposed algorithms can detect all three types of attacks with high accuracy.