Presence Of Occlusion Research Articles

IVA-former: invisible–visible query guided amodal mask measurement network for desktop object via hierarchical transformer

Abstract Instance segmentation of desktop objects is important for service robots. Most of the previous works for desktop environments are restricted to measuring the visible area of target objects. However, when a target object is placed behind another, the algorithm that only performs visible area segmentation is unable to provide accurate appearance information for the occluded objects. To solve this problem, we propose the invisible–visible query guided amodal mask measurement network based on a hierarchical transformer for desktop scenes, which can perceive the entire appearance of objects in the presence of occlusions. In this method, the RGB-D backbone is adopted to fuse the features from both RGB and depth images. Then, the pixel decoder is used to generate multi-scale feature maps. We then adopt a hierarchical transformer decoder to predict invisible, visible, and amodal masks simultaneously. To enhance the associations between the three prediction branches, we propose a query transform module to transfer object queries between adjacent branches. Since amodal masks are a combination of invisible and visible masks, we propose an invisible–visible mixture loss that takes masks from both invisible and visible branches to further supervise the network. Our method is trained on synthetic datasets for desktop objects and evaluated on both visible and amodal real-world datasets. Compared to other recent segmentation algorithms, our method achieves competitive performance.

Exploring Machine Learning and Deep Learning Techniques for Occluded Face Recognition: A Comprehensive Survey and Comparative Analysis

Face recognition occluded by occlusions, such as glasses or shadows, remains a challenge in many security and surveillance applications. This study aims to analyze the performance of various machine learning and deep learning techniques in face recognition scenarios with occlusions. We evaluate KNN (standard and FisherFace), CNN, DenseNet, Inception, and FaceNet methods combined with a pre-trained DeepFace model using three public datasets: YALE, Essex Grimace, and Georgia Tech. The results show that KNN maintains the highest accuracy, reaching 100% on two datasets (Essex Grimace and YALE), even in the presence of occlusions. Meanwhile, CNN shows strong performance, with accuracy remaining 100% on YALE, both with and without occlusions, although its performance drops slightly on Essex Grimace (94% with occlusion). DenseNet and Inception show a more significant drop in accuracy when faced with occlusion, with DenseNet dropping from 81% to 72% on Essex Grimace and Inception dropping from 100% to 92% on the same dataset. FaceNet + DeepFace excels on more large dataset (Georgia Tech) with 98% accuracy, but its performance drops dramatically to 53% and 70% on Essex Grimace and YALE with occlusion. These findings indicate that while deep learning methods show high accuracy under ideal conditions, machine learning methods such as KNN are more flexible and robust to occlusion in face recognition.

Enhancing Inter-AUV Perception: Adaptive 6-DOF Pose Estimation with Synthetic Images for AUV Swarm Sensing

The capabilities of AUV mutual perception and localization are crucial for the development of AUV swarm systems. We propose the AUV6D model, a synthetic image-based approach to enhance inter-AUV perception through 6D pose estimation. Due to the challenge of acquiring accurate 6D pose data, a dataset of simulated underwater images with precise pose labels was generated using Unity3D. Mask-CycleGAN technology was introduced to transform these simulated images into realistic synthetic images, addressing the scarcity of available underwater data. Furthermore, the Color Intermediate Domain Mapping strategy is proposed to ensure alignment across different image styles at pixel and feature levels, enhancing the adaptability of the pose estimation model. Additionally, the Salient Keypoint Vector Voting Mechanism was developed to improve the accuracy and robustness of underwater pose estimation, enabling precise localization even in the presence of occlusions. The experimental results demonstrated that our AUV6D model achieved millimeter-level localization precision and pose estimation errors within five degrees, showing exceptional performance in complex underwater environments. Navigation experiments with two AUVs further verified the model’s reliability for mutual 6D pose estimation. This research provides substantial technical support for more complex and precise collaborative operations for AUV swarms in the future.

A Multi-Scale Graph Attention-Based Transformer for Occluded Person Re-Identification

The objective of person re-identification (ReID) tasks is to match a specific individual across different times, locations, or camera viewpoints. The prevalent issue of occlusion in real-world scenarios affects image information, rendering the affected features unreliable. The difficulty and core challenge lie in how to effectively discern and extract visual features from human images under various complex conditions, including cluttered backgrounds, diverse postures, and the presence of occlusions. Some works have employed pose estimation or human key point detection to construct graph-structured information to counteract the effects of occlusions. However, this approach introduces new noise due to issues such as the invisibility of key points. Our proposed module, in contrast, does not require the use of additional feature extractors. Our module employs multi-scale graph attention for the reweighting of feature importance. This allows features to concentrate on areas genuinely pertinent to the re-identification task, thereby enhancing the model’s robustness against occlusions. To address these problems, a model that employs multi-scale graph attention to reweight the importance of features is proposed in this study, significantly enhancing the model’s robustness against occlusions. Our experimental results demonstrate that, compared to baseline models, the method proposed herein achieves a notable improvement in mAP on occluded datasets, with increases of 0.5%, 31.5%, and 12.3% in mAP scores.

L1-Tree: A novel algorithm for constructing 3D tree models and estimating branch architectural traits using terrestrial laser scanning data

Branch architecture provides crucial information for the understanding of plant trait variability and the adaptive strategies employed by trees in response to their environment. High-fidelity terrestrial laser scanning (TLS) data provide an accurate, efficient, and non-destructive means for constructing three-dimensional (3D) tree models and estimating architectural traits. However, the complex canopy structure of trees in natural forests and the presence of occlusion in TLS data pose significant challenges to achieving this goal. In this study, we present a novel algorithm, L1-Tree, for the construction of 3D tree models and the estimation of architectural traits from TLS data. This algorithm is grounded in the L1-Median algorithm and integrates a tree skeleton optimization procedure that considers the structural characteristics of tree branches. By comparing modeling results and manually derived branch traits for 24 trees of 24 species, we found that the L1-Tree algorithm achieved precision, recall, and F-score values of 0.94 for branch identification, coefficient of determination, root-mean-squared error, and normalized root-mean-squared error of 0.998, 0.068 m, and 0.3 % for branch length estimation, and a respective value of 0.958, 0.257 cm and 0.9 % for branch radius estimation. Additionally, the branch identification accuracy and accuracy in branch architectural trait estimation remained satisfactory across branch orders. Compared to established 3D tree model construction algorithms (e.g., TreeQSM), our L1-Tree algorithm demonstrated a superior capability in handling noisy environments and data gaps, making it a robust tool for TLS data-based tree architecture studies. Leaf-wood separation emerged as a crucial step influencing the performance of the L1-Tree algorithm. We observed significant drop in branch identification accuracy when using an automatic leaf-wood separation algorithm as input, highlighting the urgent need to develop effective leaf-wood separation algorithms to generate high-quality wood point clouds for tree architecture studies.

Baseline vulnerable road user injury risk in multiple U.S. dense urban driving environments

Objective Understanding and modeling baseline driving safety risk in dense urban areas represents a crucial starting point for automated driving system (ADS) safety impact analysis. The purpose of this study was to leverage naturalistic vulnerable road user (VRU) collision data to quantify collision rates, crash severity, and injury risk distributions in the absence of objective injury outcome data. Methods From over 500 million vehicle miles traveled, a total of 335 collision events involving VRUs were video verified and reconstructed (126 pedestrians, 144 cyclists, and 65 motorcyclists). Data consisted of anonymized video and sensor data (Global Positioning System and accelerometer) from vehicles equipped with Nexar dash cameras. Each event was qualitatively evaluated to assess the collision geometries and other extrinsic collision factors (e.g., time of day, roadway location, presence of occlusions). Previously published injury risk models were utilized to provide estimates of Maximum Abbreviated Injury Scale MAIS2+ and MAIS3+ injury potential. Collision severity distributions and aggregated injury risk estimates were compared. Results Vehicle speeds at the time of impact ranged from 0 to 39 mph, and VRU speeds ranged from 0 to 27 mph. In general, vehicles were traveling at lower speeds at the time of impact when turning in comparison to straight travel. Nearly all events (95%) were associated with MAIS3+ injury risk estimates below 10%. Collisions involving a potential occlusion or the vehicle responding to surprising behavior by the VRU were associated with higher estimates of injury risk than those without occlusion or with the vehicle initiating the conflict. Based on accumulated risk for each event, it can be estimated that approximately 55 persons could be moderately injured (MAIS2+) and approximately 6 persons could be seriously injured (MAIS3+). Conclusion These data indicate that responding to surprising VRU behavior, having visibility be potentially occluded, and vehicle travel behavior were associated with differences in collision speed and injury risk estimation. The specific comparisons made in this study are not intended to be comprehensive but serve as a starting point for considering baseline driving risk associated with VRU collisions in dense urban areas.

Object and Event Detection Pipeline for Rink Hockey Games

All types of sports are potential application scenarios for automatic and real-time visual object and event detection. In rink hockey, the popular roller skate variant of team hockey, it is of great interest to automatically track player movements, positions, and sticks, and also to make other judgments, such as being able to locate the ball. In this work, we present a real-time pipeline consisting of an object detection model specifically designed for rink hockey games, followed by a knowledge-based event detection module. Even in the presence of occlusions and fast movements, our deep learning object detection model effectively identifies and tracks important visual elements in real time, such as: ball, players, sticks, referees, crowd, goalkeeper, and goal. Using a curated dataset consisting of a collection of rink hockey videos containing 2525 annotated frames, we trained and evaluated the algorithm’s performance and compared it to state-of-the-art object detection techniques. Our object detection model, based on YOLOv7, presents a global accuracy of 80% and, according to our results, good performance in terms of accuracy and speed, making it a good choice for rink hockey applications. In our initial tests, the event detection module successfully detected an important event type in rink hockey games, namely, the occurrence of penalties.

Novel Illumination-invariant Face Recognition Approach via Reflectance-luminance and Local Matching Model with Weighted Voting System

In this study, a novel approach has been introduced for face recognition that is unaffected by changes in illumination. This method is based on the reflectance-luminance model and incorporates local matching using a weighted voting technique to eliminate any artifacts present in the retina images. A total of 37 different linear and nonlinear filters were tested, including high pass and low pass filters, to achieve the reflectance component of the image, which remains invariant to changes in illumination. Among these filters, the maximum filter, which is a simple filter with low computational complexity, yielded the best results in extracting the illumination invariants. It was observed that the illumination invariants obtained through this method outperformed other methods such as QI, SQI, and image enhancement techniques in terms of recognition accuracy. Importantly, the proposed method does not require any prior knowledge about the facial shape or illumination conditions and can be applied to individual images independently. Unlike many existing methods, this approach does not rely on multiple images during the training stage and does not require any parameter selection to generate the illumination invariants. To further enhance the robustness of illumination, a weighted voting system was introduced. Certain regions of the image, which may adversely affect the recognition outcome due to poor illumination, occlusion, noise, or lack of distinctive information, were identified using predefined factors such as grayscale mean, image entropy, and mutual information. The proposed method was also compared to other face recognition methods in the presence of occlusions, and it demonstrated promising results outperforming existing methods. The Python algorithm successfully detects obstructed faces, genders, and ages in videos with a face matching accuracy between 80.9% to 96.9% based on proximity.

Vision-based collective motion: A locust-inspired reductionist model.

Naturally occurring collective motion is a fascinating phenomenon in which swarming individuals aggregate and coordinate their motion. Many theoretical models of swarming assume idealized, perfect perceptual capabilities, and ignore the underlying perception processes, particularly for agents relying on visual perception. Specifically, biological vision in many swarming animals, such as locusts, utilizes monocular non-stereoscopic vision, which prevents perfect acquisition of distances and velocities. Moreover, swarming peers can visually occlude each other, further introducing estimation errors. In this study, we explore necessary conditions for the emergence of ordered collective motion under restricted conditions, using non-stereoscopic, monocular vision. We present a model of vision-based collective motion for locust-like agents: elongated shape, omni-directional visual sensor parallel to the horizontal plane, and lacking stereoscopic depth perception. The model addresses (i) the non-stereoscopic estimation of distance and velocity, (ii) the presence of occlusions in the visual field. We consider and compare three strategies that an agent may use to interpret partially-occluded visual information at the cost of the computational complexity required for the visual perception processes. Computer-simulated experiments conducted in various geometrical environments (toroidal, corridor, and ring-shaped arenas) demonstrate that the models can result in an ordered or near-ordered state. At the same time, they differ in the rate at which order is achieved. Moreover, the results are sensitive to the elongation of the agents. Experiments in geometrically constrained environments reveal differences between the models and elucidate possible tradeoffs in using them to control swarming agents. These suggest avenues for further study in biology and robotics.

Robust, Resilient Enhanced CAMSHIFT Model: Advancing Face Detection and Tracking Stability in Challenging Environments

This paper presents the development of a robust CAMSHIFT model for theoretical face detection and tracking. The proposed model integrates innovative techniques such as Perceptual Grouping, three Connected Component Operators, Weighted Adaptive Colour Histogram, and Selective Adaptation. Experimental results highlight its superior performance across scenarios like occlusions, varying illumination, near/far face tracking, skin-like background tracking, and disturbance from multiple faces. The normalized log-likelihood index serves as a robust indicator for face tracking analysis. Connected Component operations provide strong markers for error detection in video sequences. The enhanced CAMSHIFT algorithm exhibits resilience and stability, even in the presence of occlusions. Comparisons with the original CAMSHIFT reveal the enhanced model's superiority, extending tracking range to 500 cm, a calculated enhancement of 42.9 percent improvement. The study consistently favours the robust and resilient CAMSHIFT model in tracking against skin-like backgrounds and disturbances. Despite webcam convenience in used for algorithm development, the benefits of high-performance camera systems are envisioned for future research. This model is a significant advancement in face detection methods, promising improved adaptability and tracking capabilities.

BGaitR-Net: An effective neural model for occlusion reconstruction in gait sequences by exploiting the key pose information

Gait recognition in the presence of occlusion is a challenging problem and the solutions proposed to date either lack robustness or depend on several unrealistic constraints. In this work, we propose a Deep Learning framework to detect and reconstruct the occluded frames in a gait sequence. Initially, occlusion detection is done using a VGG-16 network and for each frame the corresponding pose information is represented as a one-hot encoded vector. This vector is next fused with the corresponding spatial information using a Conditional Variational Autoencoder (CVAE) to obtain an effective embedding. Following this, a Bi-directional Long Short Term Memory (Bi-LSTM) is used to predict the occluded frames using the encoded vector sequence. A decoder next transforms these predicted frames back to the image space. Our proposed reconstruction model termed the Bidirectional Gait Reconstruction Network (BGaitR-Net) is formed by stacking the CVAE, Bi-LSTM, and the decoder. The CASIA-B and OU-ISIR LP datasets are used to prepare extensive gallery sets to train each of the above sub-networks and testing is done using synthetically occluded sequences from the CASIA-B data and real-occluded sequences from the TUM-IITKGP data. A thorough evaluation of our work through Dice Score and GEINet-based recognition accuracy for varying degrees of occlusion highlight the effectiveness of our model in generating frames consistent with the temporal gait pattern. Comparative study with other existing gait recognition techniques (with or without occlusion handling mechanism) and with recent Deep Learning-based video frame prediction methods emphasizes the superiority of BGaitR-Net over the others.

Addressing Challenges of Incorporating Appearance Cues Into Heuristic Multi-Object Tracker via a Novel Feature Paradigm.

In the field of Multi-Object Tracking (MOT), the incorporation of appearance cues into tracking-by-detection heuristic trackers using re-identification (ReID) features has posed limitations on its advancement. The existing ReID paradigm involves the extraction of coarse-grained object-level feature vectors from cropped objects at a fixed input size using a ReID model, and similarity computation through a simple normalized inner product. However, MOT requires fine-grained features from different object regions and more accurate similarity measurements to identify individuals, especially in the presence of occlusion. To address these limitations, we propose a novel feature paradigm. In this paradigm, we extract the feature map from the entire frame image to preserve object sizes and represent objects using a set of fine-grained features from different object regions. These features are sampled from adaptive patches within the object bounding box on the feature map to effectively capture local appearance cues. We introduce Mutual Ratio Similarity (MRS) to accurately measure the similarity of the most discriminative region between two objects based on the sampled patches, which proves effective in handling occlusion. Moreover, we propose absolute Intersection over Union (AIoU) to consider object sizes in feature cost computation. We integrate our paradigm with advanced motion techniques to develop a heuristic Motion-Feature joint multi-object tracker, MoFe. Within it, we reformulate the track state transition of tracklets to better model their life cycle, and firstly introduce a runtime recorder after MoFe to refine trajectories. Extensive experiments on five benchmarks, i.e., GMOT-40, BDD100k, DanceTrack, MOT17, and MOT20, demonstrate that MoFe achieves state-of-the-art performance in robustness and generalizability without any fine-tuning, and even surpasses the performance of fine-tuned ReID features.

6D Pose Estimation on Point Cloud Data through Prior Knowledge Integration: A Case Study in Autonomous Disassembly

The accurate estimation of 6D pose remains a challenging task within the computer vision domain, even when utilizing 3D point cloud data. Conversely, in the manufacturing domain, instances arise where leveraging prior knowledge can yield advancements in this endeavor. This study focuses on the disassembly of starter motors to augment the engineering of product life cycles. A pivotal objective in this context involves the identification and 6D pose estimation of bolts affixed to the motors, facilitating automated disassembly within the manufacturing workflow. Complicating matters, the presence of occlusions and the limitations of single-view data acquisition, notably when motors are placed in a clamping system, obscure certain portions and render some bolts imperceptible. Consequently, the development of a comprehensive pipeline capable of acquiring complete bolt information is imperative to avoid oversight in bolt detection. In this paper, employing the task of bolt detection within the scope of our project as a pertinent use case, we introduce a meticulously devised pipeline. This multi-stage pipeline effectively captures the 6D information with regard to all bolts on the motor, thereby showcasing the effective utilization of prior knowledge in handling this challenging task. The proposed methodology not only contributes to the field of 6D pose estimation but also underscores the viability of integrating domain-specific insights to tackle complex problems in manufacturing and automation.

Design Analysis of Portable 1 Channel Infusion Device Analyzer Using Sensor SKU 237545

An infusion pump is a tool used to inject a certain amount of fluid into the patient's body through the patient's veins continuously over a certain period of time. A syringe pump is a tool that functions to push the syringe rod so that it can produce a flow ranging from microliters to milliliters per minute periodically with high accuracy. Very often there are problems with blockages or occlusion when using infusion pumps and syringe pumps. The occlusion limit set is ≤20 PSI according to ECRI. The presence of occlusion in the infusion pump and syringe pump can be identified when there is an alarm buzzer which will sound when a blockage is detected. A 1 Channel Portable Infusion Device Analyzer has been designed using the SKU 237545 Sensor, namely by using a 1 channel flowrate and occlusion sensor and making it portable to be efficient. For this reason, it is necessary to analyze the performance of the tools that have been created. How accurate is it? From the results of performance testing, Oclusion was corrected at 0.242 psi and 0.3 Psi. For flow rate, the largest correction was 2.4 ml/hour and the uncertainty was 6,046 ml/hour. This shows that the accuracy of the design is still quite high and the resulting tool is still not stable, this can be seen from the uncertainty value. The uncertainty that occurs is likely due to the sensitivity of the droplet sensor related to the detection time of the droplet

End to end stroke triage using cerebrovascular morphology and machine learning.

Rapid and accurate triage of acute ischemic stroke (AIS) is essential for early revascularization and improved patient outcomes. Response to acute reperfusion therapies varies significantly based on patient-specific cerebrovascular anatomy that governs cerebral blood flow. We present an end-to-end machine learning approach for automatic stroke triage. Employing a validated convolutional neural network (CNN) segmentation model for image processing, we extract each patient's cerebrovasculature and its morphological features from baseline non-invasive angiography scans. These features are used to detect occlusion's presence and the site automatically, and for the first time, to estimate collateral circulation without manual intervention. We then use the extracted cerebrovascular features along with commonly used clinical and imaging parameters to predict the 90 days functional outcome for each patient. The CNN model achieved a segmentation accuracy of 94% based on the Dice similarity coefficient (DSC). The automatic stroke detection algorithm had a sensitivity and specificity of 92% and 94%, respectively. The models for occlusion site detection and automatic collateral grading reached 96% and 87.2% accuracy, respectively. Incorporating the automatically extracted cerebrovascular features significantly improved the 90 days outcome prediction accuracy from 0.63 to 0.83. The fast, automatic, and comprehensive model presented here can improve stroke diagnosis, aid collateral assessment, and enhance prognostication for treatment decisions, using cerebrovascular morphology.

A Robust Lightweight Network for Pedestrian Detection Based on YOLOv5-x

Pedestrian detection is a crucial task in computer vision, with various applications in surveillance, autonomous driving, and robotics. However, detecting pedestrians in complex scenarios, such as rainy days, remains a challenging problem due to the degradation of image quality and the presence of occlusions. To address this issue, we propose RSTDet-Lite (a robust lightweight network) for pedestrian detection on rainy days, based on an improved version of YOLOv5-x. Specifically, in order to reduce the redundant parameters of the YOLOv5-x backbone network and enhance its feature extraction capability, we propose a novel approach named CBP-GNet, which incorporates a compact bilinear pooling algorithm. This new net serves as a new backbone network, resulting in significant parameter reduction and enhancing the fine-grained feature fusion capability of the network. Additionally, we introduce the Simple-BiFPN structure as a replacement for the original feature pyramid module based on the weighted bidirectional feature pyramid to further improve feature fusion efficiency. To enhance network performance, we integrate the CBAM attention mechanism and introduce the idea of structural reparameterization. To evaluate the performance of our method, we create a new dataset named RainDet3000, which consists of 3000 images captured in various rainy scenarios. The experimental results demonstrate that, compared with YOLOv5, our proposed model reduces the network size by 30 M while achieving a 4.56% increase in mAP. This confirms the effectiveness of RSTDet-Lite in achieving excellent performance in rainy-day pedestrian detection scenarios.

Aphid Detection Network from Global to Local

Common aphids on crops are not suitable for general-purpose object detection frameworks due to their small size and the presence of occlusion. Taking this as a starting point, we fully consider the characteristics of aphid targets and propose a network called Overall-Specific Net (OS-Net) for detecting aphid targets in a manner that goes from overall to specific in a single-stage detection network. This network consists of two detection head modules. The first detection head module is responsible for detecting aphids in densely populated areas, while the second module, based on the first module, detects aphids in densely populated areas by deploying denser anchor boxes. Experimental results on our dataset show that the average accuracy can be improved by approximately 5.1% compared to the baseline network.

Research on automatic classification and detection of chicken parts based on deep learning algorithm.

Accurate classification and identification of chicken parts are critical to improve the productivity and processing speed in poultry processing plants. However, the overlapping of chicken parts has an impact on the effectiveness of the identification process. To solve this issue, this study proposed a real-time classification and detection method for chicken parts, utilizing YOLOV4 deep learning. The method can identify segmented chicken parts on the assembly line in real time and accurately, thus improving the efficiency of poultry processing. First, 600 images containing multiple chicken part samples were collected to build a chicken part dataset after using the image broadening technique, and then the dataset was divided according to the 6:2:2 division principle, with 1200 images as the training set, 400 images as the test set, and 400 images as the validation set. Second, we utilized the single-stage target detector YOLO to predict and calculate the chicken part images, obtaining the categories and positions of the chicken leg, chicken wing, and chicken breast in the image. This allowed us to achieve real-time classification and detection of chicken parts. This approach enabled real-time and efficient classification and detection of chicken parts. Finally, the mean average precision (mAP) and the processing time per image were utilized as key metrics to evaluate the effectiveness of the model. In addition, four other target detection algorithms were introduced for comparison with YOLOV4-CSPDarknet53 in this study, which include YOLOV3-Darknet53, YOLOV3-MobileNetv3, SSD-MobileNetv3, and SSD-VGG16. A comprehensive comparison test was conducted to assess the classification and detection performance of these models for chicken parts. Finally, for the chicken part dataset, the mAP of the YOLOV4-CSPDarknet53 model was 98.86% on a single image with an inference speed of 22.2 ms, which was higher than the other four models of YOLOV3-Darknet53, YOLOV3-MobileNetv3, SSD-MobileNetv3, and SSD-VGG16 mAP by 3.27%, 3.78%, 6.91%, and 6.13%, respectively. The average detection time was reduced by 13, 1.9, 6.2, and 20.3 ms, respectively. In summary, the chicken part classification and detection method proposed in this study offers numerous benefits, including the ability to detect multiple chicken parts simultaneously, as well as delivering high levels of accuracy and speed. Furthermore, this approach effectively addresses the issue of accurately identifying individual chicken parts in the presence of occlusion, thereby reducing waste on the assembly line. PRACTICAL APPLICATION: The aim of this study is to offer visual technical assistance in minimizing wastage and resource depletion during the sorting and cutting of chicken parts in poultry production and processing facilities. Furthermore, considering the diverse demands and preferences regarding chicken parts, this research can facilitate product processing that caters specifically to consumer preferences.

Efficient Annotation and Learning for 3D Hand Pose Estimation: A Survey

AbstractIn this survey, we present a systematic review of 3D hand pose estimation from the perspective of efficient annotation and learning. 3D hand pose estimation has been an important research area owing to its potential to enable various applications, such as video understanding, AR/VR, and robotics. However, the performance of models is tied to the quality and quantity of annotated 3D hand poses. Under the status quo, acquiring such annotated 3D hand poses is challenging, e.g., due to the difficulty of 3D annotation and the presence of occlusion. To reveal this problem, we review the pros and cons of existing annotation methods classified as manual, synthetic-model-based, hand-sensor-based, and computational approaches. Additionally, we examine methods for learning 3D hand poses when annotated data are scarce, including self-supervised pretraining, semi-supervised learning, and domain adaptation. Based on the study of efficient annotation and learning, we further discuss limitations and possible future directions in this field.

Leaf counting in the presence of occlusion in Arabidopsis thaliana plant using convolutional neural networks

Leaf counting in the presence of occlusion in Arabidopsis thaliana plant using convolutional neural networks