Bounding Box Research Articles

More Is Not Always Better: Impacts of AI-Generated Confidence and Explanations in Human-Automation Interaction.

The study aimed to enhance transparency in autonomous systems by automatically generating and visualizing confidence and explanations and assessing their impacts on performance, trust, preference, and eye-tracking behaviors in human-automation interaction. System transparency is vital to maintaining appropriate levels of trust and mission success. Previous studies presented mixed results regarding the impact of displaying likelihood information and explanations, and often relied on hand-created information, limiting scalability and failing to address real-world dynamics. We conducted a dual-task experiment involving 42 university students who operated a simulated surveillance testbed with assistance from intelligent detectors. The study used a 2 (confidence visualization: yes vs. no) × 3 (visual explanations: none, bounding boxes, bounding boxes and keypoints) mixed design. Task performance, human trust, preference for intelligent detectors, and eye-tracking behaviors were evaluated. Visual explanations using bounding boxes and keypoints improved detection task performance when confidence was not displayed. Meanwhile, visual explanations enhanced trust and preference for the intelligent detector, regardless of the explanation type. Confidence visualization did not influence human trust in and preference for the intelligent detector. Moreover, both visual information slowed saccade velocities. The study demonstrated that visual explanations could improve performance, trust, and preference in human-automation interaction without confidence visualization partially by changing the search strategies. However, excessive information might cause adverse effects. These findings provide guidance for the design of transparent automation, emphasizing the importance of context-appropriate and user-centered explanations to foster effective human-machine collaboration.

Learning to Holistically Detect Bridges From Large-Size VHR Remote Sensing Imagery.

Bridge detection in remote sensing images (RSIs) plays a crucial role in various applications, but it poses unique challenges compared to the detection of other objects. In RSIs, bridges exhibit considerable variations in terms of their spatial scales and aspect ratios. Therefore, to ensure the visibility and integrity of bridges, it is essential to perform holistic bridge detection in large-size very-high-resolution (VHR) RSIs. However, the lack of datasets with large-size VHR RSIs limits the deep learning algorithms' performance on bridge detection. Due to the limitation of GPU memory in tackling large-size images, deep learning-based object detection methods commonly adopt the cropping strategy, which inevitably results in label fragmentation and discontinuous prediction. To ameliorate the scarcity of datasets, this paper proposes a large-scale dataset named GLH-Bridge comprising 6,000 VHR RSIs sampled from diverse geographic locations across the globe. These images encompass a wide range of sizes, varying from 2,048 × 2,048 to 16,384 × 16,384 pixels, and collectively feature 59,737 bridges. These bridges span diverse backgrounds, and each of them has been manually annotated, using both an oriented bounding box (OBB) and a horizontal bounding box (HBB). Furthermore, we present an efficient network for holistic bridge detection (HBD-Net) in large-size RSIs. The HBD-Net presents a separate detector-based feature fusion (SDFF) architecture and is optimized via a shape-sensitive sample re-weighting (SSRW) strategy. The SDFF architecture performs inter-layer feature fusion (IFF) to incorporate multi-scale context in the dynamic image pyramid (DIP) of the large-size image, and the SSRW strategy is employed to ensure an equitable balance in the regression weight of bridges with various aspect ratios. Based on the proposed GLH-Bridge dataset, we establish a bridge detection benchmark including the OBB and HBB tasks, and validate the effectiveness of the proposed HBD-Net. Additionally, cross-dataset generalization experiments on two publicly available datasets illustrate the strong generalization capability of the GLH-Bridge dataset.

RepSGG: Novel Representations of Entities and Relationships for Scene Graph Generation.

Scene Graph Generation (SGG) has achieved significant progress recently. However, most previous works rely heavily on fixed-size entity representations based on bounding box proposals, anchors, or learnable queries. As each representation's cardinality has different trade-offs between performance and computation overhead, extracting highly representative features efficiently and dynamically is both challenging and crucial for SGG. In this work, a novel architecture called RepSGG is proposed to address the aforementioned challenges, formulating a subject as queries, an object as keys, and their relationship as the maximum attention weight between pairwise queries and keys. With more fine-grained and flexible representation power for entities and relationships, RepSGG learns to sample semantically discriminative and representative points for relationship inference. Moreover, the long-tailed distribution also poses a significant challenge for generalization of SGG. A run-time performance-guided logit adjustment (PGLA) strategy is proposed such that the relationship logits are modified via affine transformations based on run-time performance during training. This strategy encourages a more balanced performance between dominant and rare classes. Experimental results show that RepSGG achieves the state-of-the-art or comparable performance on the Visual Genome and Open Images V6 datasets with fast inference speed, demonstrating the efficacy and efficiency of the proposed methods.

Laser weld spot detection based on YOLO-weld.

Laser weld point detection is crucial in modern industrial manufacturing, yet it faces challenges such as a limited number of samples, uneven distribution, and diverse, irregular shapes. To address these issues, this paper proposes an innovative model, YOLO-Weld, which achieves lightweight design while enhancing detection accuracy. Firstly, a targeted data augmentation strategy is employed to increase both the quantity and diversity of samples from minority classes. Following this, a Diverse Class Normalization Loss (DCNLoss)function is designed to emphasize the importance of tail data in the model's training. Secondly, the Adaptive Hierarchical Intersection over Union Loss (AHIoU Loss)function is introduced, which assigns varying levels of attention to different Intersections over Union (IoU) samples, with a particular focus on moderate IoU samples, thereby accelerating the bounding box regression process. Finally, a lightweight multi-scale feature processing module, MSBCSPELAN, is proposed to enhance multi-scale feature handling while reducing the number of model parameters. Experimental results indicate that YOLO-Weld significantly improves the accuracy and efficiency of laser weld point detection, with mean Average Precision at 50 ([Formula: see text]) and mean Average Precision at 50:95 ([Formula: see text]) increasing by 15.6% and 15.8%, respectively. Additionally, the model's parameter count is reduced by 0.4M, GFLOPS decreases by 1.1, precision improves by 4.3%, recall rises by 22.2%, and the F1 score increases by 15.1%.

Safety detection of power inspection personnel based on context decoupling and bounding box similarity measurements

Safety detection of power inspection personnel based on context decoupling and bounding box similarity measurements

Fish-Finder: A robust small target detection method for aquaculture fish in low-quality underwater images.

Underwater fish object detection serves as a pivotal research direction in marine biology, aquaculture management, and computer vision, yet it poses substantial challenges due to the complexity of underwater environments, occultations, and the small-sized and frequently moving fish in aquaculture. Addressing these challenges, we propose a novel underwater fish object detection algorithm named Fish-Finder. First, we engendered a structure titled "C2fBF," utilizing the dual-path routing attention protocol of BiFormer. The primary objective of this structure is to alleviate the perturbations induced by underwater intricacies during the phase of downsampling in the backbone network, thereby discerning and conserving finer contextual features. Subsequently, we co-opted the RepGFPN method within our neck network-a distinctive approach that adeptly merges high-level semantic constructs with low-level spatial specifics, thus fortifying its multi-scale detection prowess. Then, in an endeavor to diminish the sensitivity toward positional aberrations during the detection of diminutive aquatic creatures, we incorporated a novel bounding box regression loss function, the Wasserstein loss, to the existing CIoU. This innovative function gauges the congruity between the predicted bounding box Gaussian distribution and the reference bounding box Gaussian distribution. Finally, in regard to the dataset, we independently assembled a specific dataset termed "SmallFish." This unique dataset, meticulously designed for the detection of small-scale fish within intricate underwater settings, includes 5000 annotated images of small fish. Experimental results demonstrate that, compared to the state-of-the-art detection methods, our proposed method improves the accuracy by and , and mean average precision (mAP) increases and in public dataset Kaggle-Fish and our SmallFish dataset, respectively.

Synthetizing Virtual Construction Images to Strengthen Real Data Volume and Variety in Real-world Application Scenarios

Despite the potential of synthetic construction images, it remains unknown whether they can strengthen real-data volume and variety in real-world scenarios, wherein a given, real training dataset is small and biased, or large but biased. To address this, we synthetize artificial images in a computer environment to strengthen a real training dataset and test its supplementary effects in both scenarios. Specifically, we simulate a worker’s physical behaviors, capture 2D synthetic images, and annotate its bounding box using a 3D-2D projection algorithm. After combining these synthetic images with a real dataset, we train a vision-based worker detection model and evaluate its performance in each scenario. Results show that the model’s performance is improved by up to 59.1% and 12.8% in each scenario, respectively, comparing to only adopting real images. This indicates that synthetic images can enrich the restricted volume and variety of a given, real training dataset in field application scenarios.

Distinguishing Difficulty Imbalances in Strawberry Ripeness Instances in a Complex Farmland Environment

The existing strawberry ripeness detection algorithm has the problems of a low precision and a high missing rate in real complex scenes. Therefore, we propose a novel model based on a hybrid attention mechanism. Firstly, a partial convolution-based compact inverted block is developed, which significantly enhances the feature extraction capability of the model. Secondly, an efficient partial hybrid attention mechanism is established, which realizes the remote dependence and accurate localization of strawberry fruit. Meanwhile, a multi-scale progressive feature pyramid network is constructed, and the fine-grained features of strawberry targets of different sizes are accurately extracted. Finally, a Focaler-shape-IoU loss function is proposed to effectively solve the problem of the difficulty imbalance between strawberry samples and the influence of the shape and size of the bounding box on the regression. The experimental results show that the model’s precision and mAP0.5 reach 92.1% and 92.7%, respectively, which are 2.0% and 1.7% higher than the baseline model. Additionally, our model is better in detection performance than most models with fewer parameters and lower FLOPs. In summary, the model can accurately identify the maturity of strawberry fruit under complex farmland environments and provide certain technical guidance for automated strawberry-picking robots.

Lightweight Neural Network for Centroid Detection of Weak, Small Infrared Targets via Background Matching in Complex Scenes

In applications such as aerial object interception and ballistic estimation, it is crucial to precisely detect the centroid position of the target rather than to merely identify the position of the target bounding box or segment all pixels belonging to the target. Due to the typically long distances between targets and imaging devices in such scenarios, targets often exhibit a low contrast and appear as dim, obscure shapes in infrared images, which represents a challenge for human observation. To rapidly and accurately detect small targets, this paper proposes a lightweight, end-to-end detection network for small infrared targets. Unlike existing methods, the input of this network is five consecutive images after background matching. This design significantly improves the network’s ability to extract target motion features and effectively reduces the interference of static backgrounds. The network mainly consists of a local feature aggregation module (LFAM), which uses multiple-sized convolution kernels to capture multi-scale features in parallel and integrates multiple spatial attention mechanisms to achieve accurate feature fusion and effective background suppression, thereby enhancing the ability to detect small targets. To improve the accuracy of predicted target centroids, a centroid correction algorithm is designed. In summary, this paper presents a lightweight centroid detection network based on background matching for weak, small infrared targets. The experimental results show that, compared to directly inputting a sequence of images into the neural network, inputting a sequence of images processed by background matching can increase the detection rate by 9.88%. Using the centroid correction algorithm proposed in this paper can therefore improve the centroid localization accuracy by 0.0134.

Ambiguous facial expression detection for Autism Screening using enhanced YOLOv7-tiny model

Autism spectrum disorder is a developmental condition that affects the social and behavioral abilities of growing children. Early detection of autism spectrum disorder can help children to improve their cognitive abilities and quality of life. The research in the area of autism spectrum disorder reports that it can be detected from cognitive tests and physical activities of children. The present research reports on the detection of autism spectrum disorder from the facial attributes of children. Children with autism spectrum disorder show ambiguous facial expressions which are different from the facial attributes of normal children. To detect autism spectrum disorder from facial images, this work presents an improvised variant of the YOLOv7-tiny model. The presented model is developed by integrating a pyramid of dilated convolutional layers in the feature extraction network of the YOLOv7-tiny model. Further, its recognition abilities are enhanced by incorporating an additional YOLO detection head. The developed model can detect faces with the presence of autism features by drawing bounding boxes and confidence scores. The entire work has been carried out on a self-annotated autism face dataset. The developed model achieved a mAP value of 79.56% which was better than the baseline YOLOv7-tiny and state-of-the-art YOLOv8 Small model.

FETrack: Feature-Enhanced Transformer Network for Visual Object Tracking

Visual object tracking is a fundamental task in computer vision, with applications ranging from video surveillance to autonomous driving. Despite recent advances in transformer-based one-stream trackers, unrestricted feature interactions between the template and the search region often introduce background noise into the template, degrading the tracking performance. To address this issue, we propose FETrack, a feature-enhanced transformer-based network for visual object tracking. Specifically, we incorporate an independent template stream in the encoder of the one-stream tracker to acquire the high-quality template features while suppressing the harmful background noise effectively. Then, we employ a sequence-learning-based causal transformer in the decoder to generate the bounding box autoregressively, simplifying the prediction head network. Further, we present a dynamic threshold-based online template-updating strategy and a template-filtering approach to boost tracking robustness and reduce redundant computations. Extensive experiments demonstrate that our FETrack achieves a superior performance over state-of-the-art trackers. Specifically, the proposed FETrack achieves a 75.1% AO on GOT-10k, 81.2% AUC on LaSOT, and 89.3% Pnorm on TrackingNet.

Method to Estimate Dislocation Densities from Images of α‐Ga2O3‐Based Corundum Oxides Using the Computer Vision YOLO Algorithm

This work applies the computer vision “You only look once” (YOLO) algorithm to extract bounding boxes around dislocations in weak‐beam dark‐field transmission electron microscopy (WBDF TEM) images of semiconductor thin films. A formula is derived to relate the sum of the relative heights of the bounding boxes to the dislocation densities in the films. WBDF TEM images reported in the literature and taken from our α‐Ga2O3 samples are divided into train, evaluation, and test datasets. Different models are trained using the train dataset and evaluated using the evaluation dataset to find the best confidence values, which are used to select the best model based on the performance against the test data set. For α‐Ga2O3 thin films, dislocation density output by this model is on average ≈58% of those estimated by the traditional Ham method. A factor of 4/π may contribute to the systematic underestimation of the model versus the Ham method.

Damage Detection and Segmentation in Disaster Environments Using Combined YOLO and Deeplab

Building damage due to various causes occurs frequently and has risk factors that can cause additional collapses. However, it is difficult to accurately identify objects in complex structural sites because of inaccessible situations and image noise. In conventional approaches, close-up images have been used to detect and segment damage images such as cracks. In this study, the method of using a deep learning model is proposed for the rapid determination and analysis of multiple damage types, such as cracks and concrete rubble, in disaster sites. Through the proposed method, it is possible to perform analysis by receiving image information from a robot explorer instead of a human, and it is possible to detect and segment damage information even when the damaged point is photographed at a distance. To accomplish this goal, damage information is detected and segmented using YOLOv7 and Deeplabv2. Damage information is quickly detected through YOLOv7, and semantic segmentation is performed using Deeplabv2 based on the bounding box information obtained through YOLOv7. By using images with various resolutions and senses of distance for training, damage information can be effectively detected not only at short distances but also at long distances. When comparing the results, depending on how YOLOv7 and Deeplabv2 were used, they returned better scores than the comparison model, with a Recall of 0.731, Precision of 0.843, F1 of 0.770, and mIoU of 0.638, and had the lowest standard deviation.

Insulator-YOLO: Transmission Line Insulator Risk Identification Based on Improved YOLOv5

This study introduces an innovative method for detecting risks in transmission line insulators by developing an optimized variant of YOLOv5, named Insulator-YOLO. The model addresses key challenges in small-defect detection, complex backgrounds, and computational efficiency. By incorporating GhostNetV2 in the backbone to streamline feature extraction and introducing SE and CBAM attention mechanisms, the model enhances its focus on critical features. The Bibi-directional Feature feature Pyramid pyramid Network network (BiFPN) is applied to enhance multi-scale feature fusion, and the integration of CIoU and NWD loss functions optimizes bounding box regression, achieving higher accuracy. Additionally, focal loss mitigates the imbalance between positive and negative samples, leading to more accurate and robust defect detection. Extensive evaluations demonstrate that Insulator-YOLO significantly improves detection accuracy and efficiency in real-world power line insulator defects, providing a reliable solution for maintaining the integrity of transmission systems.

Analysis of the effectiveness of using U-net architecture for classification and segmentation of glioma in MRI images

The purpose of the research is to analyze the efficiency of the U-net neural network architecture in decision support systems for glioma diagnostics and segmentation of brain areas affected by it on MRI images.Methods. To conduct experimental studies, a training dataset was generated and the data was normalized. A software implementation of the U-Net neural network architecture was performed using the Keras framework in the Python programming language. The neural network model was trained.Results. A series of experiments were conducted, during which error and classification matrices were obtained, the efficiency of classification of the trained neural network model for the "Tumor" and "No tumor" classes was assessed using metrics such as Recall, Precision and F1-measure, and the quality of segmentation of glioma-affected areas on the test data set was assessed. The quality of segmentation was assessed using the IoU metric, which reflects the ratio of the areas of the bounding boxes and is used to assess the accuracy of the spatial correspondence of the predicted segmented areas highlighted on the masks. Based on the results of testing the neural network model in solving the problem of segmenting brain areas affected by glioma, the average value of the IoU metric was 0.812, which is an acceptable result.Conclusion. The testing results showed that the neural network model based on the U-net architecture is able to effectively diagnose the presence of glioma with acceptable values of the classification and segmentation quality metrics, which indicates the possibility of using this neural network model in medical decision support systems for glioma diagnostics, as well as its segmentation on MRI images. However, it is advisable to refine this neural network model to reduce the number of false negative classification results, which is critically important in medical diagnostics.

Vision-Based Fall Detector for Elderly Based on Sliding Window Approach and Feature Engineering

A new vision-based fall detector is proposed that uses the tsfresh tool to generate features from the bounding box motion parameters of an object and performs classification in a sliding window mode. The efficiency of the generated features is demonstrated compared to the primary ones. Using the auto-sklearn library and a generalized dataset compiled from the UR Fall Detection and CAUCAFall datasets, the best human fall detection model is found. This model, based on a gradient boosting classifier, achieved 96% accuracy, which is not inferior to well-known detection algorithms, but uses only two primary motion parameters to generate secondary features. A PCA-based class separability study showed that for secondary features, 99% of the variance is captured by the first 4 principal components, while for primary features, the first 10 principal components contain only 80% of the data variance. Furthermore, the processing time for generating secondary features and making predictions was found to be relatively short, taking only a few seconds per sequence, highlighting the practical applicability of the proposed approach in real-time fall monitoring systems.

Improving Visual Object Detection using General UFIR Filtering

Object detection is a fundamental task in computer vision, which involves the identification and localization of objects within image frames or video sequences. The problem is complicated by large variations in the video camera bounding box, which can be thought of as colored measurement noise (CMN). In this paper, we use the general unbiased finite impulse response (GUFIR) approach to improve detection performance under CMN. The results are compared to the general Kalman filter (GKF) and two detection methods: “Faster-RCNN” and “Tensorflow PASCAL Visual Object Classes (VOC)”. Experimental testing is carried out using the benchmark data ”Car4”. It is shown that GUFIR significantly improves the detection accuracy and demonstrates the properties of the effective tool for visual object tracking.

A Comprehensive Framework for Transportation Infrastructure Digitalization: TJYRoad-Net for Enhanced Point Cloud Segmentation.

This research introduces a cutting-edge approach to traffic infrastructure digitization, integrating UAV oblique photography with LiDAR point clouds for high-precision, lightweight 3D road modeling. The proposed method addresses the challenge of accurately capturing the current state of infrastructure while minimizing redundancy and optimizing computational efficiency. A key innovation is the development of the TJYRoad-Net model, which achieves over 85% mIoU segmentation accuracy by including a traffic feature computing (TFC) module composed of three critical components: the Regional Coordinate Encoder (RCE), the Context-Aware Aggregation Unit (CAU), and the Hierarchical Expansion Block. Comparative analysis segments the point clouds into road and non-road categories, achieving centimeter-level registration accuracy with RANSAC and ICP. Two lightweight surface reconstruction techniques are implemented: (1) algorithmic reconstruction, which delivers a 6.3 mm elevation error at 95% confidence in complex intersections, and (2) template matching, which replaces road markings, poles, and vegetation using bounding boxes. These methods ensure accurate results with minimal memory overhead. The optimized 3D models have been successfully applied in driving simulation and traffic flow analysis, providing a practical and scalable solution for real-world infrastructure modeling and analysis. These applications demonstrate the versatility and efficiency of the proposed methods in modern traffic system simulations.

Detection and identification of centipedes based on deep learning

The quantification of centipede populations is one of the key measures in achieving intelligent management of edible centipedes and promoting the upgrade of the rural centipede industry chain. However, current centipede counting techniques still face several challenges, including low detection accuracy, large model size, and difficulty in deployment on mobile devices. These challenges have limited existing network models to the experimental stage, preventing their practical application. To tackle the identified challenges, this study introduces a lightweight centipede detection model (FCM-YOLO), which enhances detection performance while ensuring fast processing and broad applicability. Based on the YOLOv5s framework, this model incorporates the C3FS module, resulting in fewer parameters and increased detection speed. Additionally, it integrates an attention module (CBAM) to suppress irrelevant information and improve target focus, thus enhancing detection accuracy. Furthermore, to enhance the precision of bounding box positioning, this study proposes a new loss function, CMPDIOU, for bounding box loss. Experimental results show that FCM-YOLO, while reducing parameter size, achieves an improved detection accuracy of 97.4% (2.7% higher than YOLOv5s) and reduces floating-point operations (FLOPs) to 11.5G (4.3G lower than YOLOv5s). In summary, this paper provides novel insights into the detection and enumeration of centipedes, contributing to the advancement of intelligent agricultural practices.

A Quantitative Comparison Between Human and Artificial Intelligence in the Detection of Focal Cortical Dysplasia.

Artificial intelligence (AI) is thought to improve lesion detection. However, a lack of knowledge about human performance prevents a comparative evaluation of AI and an accurate assessment of its impact on clinical decision-making. The objective of this work is to quantitatively evaluate the ability of humans to detect focal cortical dysplasia (FCD), compare it to state-of-the-art AI, and determine how it may aid diagnostics. We prospectively recorded the performance of readers in detecting FCDs using single points and 3-dimensional bounding boxes. We acquired predictions of 3 AI models for the same dataset and compared these to readers. Finally, we analyzed pairwise combinations of readers and models. Twenty-eight readers, including 20 nonexpert and 5 expert physicians, reviewed 180 cases: 146 subjects with FCD (median age: 25, interquartile range: 18) and 34 healthy control subjects (median age: 43, interquartile range: 19). Nonexpert readers detected 47% (95% confidence interval [CI]: 46, 49) of FCDs, whereas experts detected 68% (95% CI: 65, 71). The 3 AI models detected 32%, 51%, and 72% of FCDs, respectively. The latter, however, also predicted more than 13 false-positive clusters per subject on average. Human performance was improved in the presence of a transmantle sign ( P < 0.001) and cortical thickening ( P < 0.001). In contrast, AI models were sensitive to abnormal gyration ( P < 0.01) or gray-white matter blurring ( P < 0.01). Compared with single experts, expert-expert pairs detected 13% (95% CI: 9, 18) more FCDs ( P < 0.001). All AI models increased expert detection rates by up to 19% (95% CI: 15, 24) ( P < 0.001). Nonexpert+AI pairs could still outperform single experts by up to 13% (95% CI: 10, 17). This study pioneers the comparative evaluation of humans and AI for FCD lesion detection. It shows that AI and human predictions differ, especially for certain MRI features of FCD, and, thus, how AI may complement the diagnostic workup.