Bounding Box Labels Research Articles

A Feasibility Study of Thermography for Detecting Pressure Injuries Across Diverse Skin Tones.

Pressure injury (PI) detection is challenging, especially in dark skin tones, due to the unreliability of visual inspection. Thermography may serve as a viable alternative as temperature differences in the skin can indicate impending tissue damage. Although deep learning models hold considerable promise toward reliably detecting PI, existing work fails to evaluate performance on diverse skin tones and varying data collection protocols. We collected a new dataset of 35 participants focused on darker skin tones where temperature differences are induced through cooling and cupping protocols. The dataset includes different cameras, lighting, patient pose, and camera distance. We compare the performance of three convolutional neural network (CNN) models trained on either the thermal or the optical images on all skin tones. Our results suggest thermography-based CNN is robust to data collection protocols. Moreover, the visual explanation often captures the region of interest without requiring explicit bounding box labels.

Leveraging Perspective Transformation for Enhanced Pothole Detection in Autonomous Vehicles.

Road conditions, often degraded by insufficient maintenance or adverse weather, significantly contribute to accidents, exacerbated by the limited human reaction time to sudden hazards like potholes. Early detection of distant potholes is crucial for timely corrective actions, such as reducing speed or avoiding obstacles, to mitigate vehicle damage and accidents. This paper introduces a novel approach that utilizes perspective transformation to enhance pothole detection at different distances, focusing particularly on distant potholes. Perspective transformation improves the visibility and clarity of potholes by virtually bringing them closer and enlarging their features, which is particularly beneficial given the fixed-size input requirement of object detection networks, typically significantly smaller than the raw image resolutions captured by cameras. Our method automatically identifies the region of interest (ROI)-the road area-and calculates the corner points to generate a perspective transformation matrix. This matrix is applied to all images and corresponding bounding box labels, enhancing the representation of potholes in the dataset. This approach significantly boosts detection performance when used with YOLOv5-small, achieving a 43% improvement in the average precision (AP) metric at intersection-over-union thresholds of 0.5 to 0.95 for single class evaluation, and notable improvements of 34%, 63%, and 194% for near, medium, and far potholes, respectively, after categorizing them based on their distance. To the best of our knowledge, this work is the first to employ perspective transformation specifically for enhancing the detection of distant potholes.

Enhancing medical image analysis with unsupervised domain adaptation approach across microscopes and magnifications

In the domain of medical image analysis, deep learning models are heralding a revolution, especially in detecting complex and nuanced features characteristic of diseases like tumors and cancers. However, the robustness and adaptability of these models across varied imaging conditions and magnifications remain a formidable challenge. This paper introduces the Fourier Adaptive Recognition System (FARS), a pioneering model primarily engineered to address adaptability in malarial parasite recognition. Yet, the foundational principles guiding FARS lend themselves seamlessly to broader applications, including tumor and cancer diagnostics. FARS capitalizes on the untapped potential of transitioning from bounding box labels to richer semantic segmentation labels, enabling a more refined examination of microscopy slides. With the integration of adversarial training and the Color Domain Aware Fourier Domain Adaptation (F2DA), the model ensures consistent feature extraction across diverse microscopy configurations. The further inclusion of category-dependent context attention amplifies FARS’s cross-domain versatility. Evidenced by a substantial elevation in cross-magnification performance from 31.3% mAP to 55.19% mAP and a 15.68% boost in cross-domain adaptability, FARS positions itself as a significant advancement in malarial parasite recognition. Furthermore, the core methodologies of FARS can serve as a blueprint for enhancing precision in other realms of medical image analysis, especially in the complex terrains of tumor and cancer imaging. The code is available at; https://github.com/Mr-TalhaIlyas/FARS.

High-speed acquisition and intelligent tunnel surface defects recognition

We introduce a framework to efficiently capture and identify the defects on the subway tunnel surfaces. Specifically, we design a high-speed multi-camera-synchronous acquisition system for scanning the tunnel surfaces and capturing the cracks from images. We propose an Improved Cascade Region-Based Convolutional Neural Network (R-CNN) method to locate and classify defects, with the crack being refined by image processing algorithms. The methodology is tested by on-site experiments carried out in an operational subway. The mAP value reaches 0.827, with a specific AP value of 0.625 and a newly proposed measure identification rate of 0.942 for crack objects. Notably, by training on bounding box labels, our overall algorithm exhibits the lowest Error Rate (ER) and the highest Intersection over Union (IoU) across all test images, surpassing the performance of classical semantic segmentation networks trained on pixel-level labels.

Framework of rod-like crops sorting based on multi-object oriented detection and analysis

Quality grading and sorting are crucial post-harvest procedures to increase the market value of crops. To reduce the over-dependence on manual work for these tasks, automated sorting methods have been studied. However, we find existing research focuses mainly on round-like fruits and vegetables, and most machines can only grade one object at a time in their vision pipeline. To address these limitations, this paper proposes a comprehensive framework specifically for detecting and analyzing rod-like crops based on multi-object oriented detection. Zizania shoots are primarily used to validate our methods. To implement the deep learning models, an efficient oriented bounding box label software called OBBLabel is developed, and several large-scale image datasets are constructed. Both of the software and datasets are open-sourced for the community. Based on YOLOv8 architecture, the proposed YOLO-OBB model predicts oriented bounding boxes to extract all rod-like targets in image with 0.903 mAP@0.5. A multi-label recognition model YOLO-MLD then conducts quality grading and posture perception on individual target with 93.4% mean accuracy. Thus, precise position and quality information for all objects can be obtained in near real-time for subsequent suction cup-based sorting operations. Furthermore, according to the framework, this paper designs and manufactures a prototype sorting machine with edge intelligence for rod-like crops.

Weakly-supervised learning based automatic augmentation of aerial insulator images

The automatic detection of insulator defects with UAV and CNN-based detectors has become a popular paradigm in recent years. However, insufficient insulator data has always been the bottleneck of detector performance. The existing augmentation method either performs a whole image transformation that lacks new semantic features or generates samples with massive manual annotation. Therefore, this paper proposes an automatic augmentation method called Weakly-Supervised Segmentation Mix (WSSM), where a Foreground Segmentation Network (FSN) is trained under the supervision of the bounding box label to extract the insulators for new sample synthesis. In the FSN training process, UnionMix is designed to generate hard samples based on the pseudo-label, thus facilitating the FSN segmentation ability for insignificant insulator boundaries. Oriented Muti-Instance Loss (OMIL) is proposed to extract supervision from the oriented bounding box so that FSN can be fully trained to handle the diverse angle distribution of insulators. The experiments conducted on the Aerial Insulator Dataset (AID) indicate that the synthesized images of WSSM can achieve stable improvement on multiple mainstream detectors. Both in-domain and cross-domain backgrounds can be used in WSSM to promote the network. For example, the AP of YOLOv5-m can be improved from 68.14 to 69.73 with 2600 COCO images, which exceeds the AP of bassline YOLOv5-l (69.69). To further verify the foreground extraction capability, this paper takes the FSN result as the pseudo-label and trains the instance segmentation network on iSAID. The comparison with existing SOTA methods proves the superior quality of the proposed method.

VoxelScape: Large Scale Simulated 3D Point Cloud Dataset of Urban Traffic Environments

Having a profound understanding of the surrounding environment is considered one of the crucial tasks for the reliable operation of future self-driving cars. Light Detection and Ranging (LiDAR) sensor plays a critical role in achieving such understanding due to its capability to perceive the world in 3D. Similar to 2D perception tasks, current state-of-the-art methods in 3D perception tasks rely on deep neural networks (DNNs). However, the performance of 3D perception tasks, specially point-wise semantic segmentation, is not on par with their 2D counterparts. One of the main reasons is the lack of publicly available labelled 3D point cloud datasets (PCDs) from 3D LiDAR sensors. In this work, we are introducing the VoxelScape dataset, a large-scale simulated 3D PCD with 100K annotated point cloud scans. The annotations in the VoxelScape dataset include both point-wise semantic labels and 3D bounding box labels. Additionally, we used a number of baseline approaches to validate the transferability of VoxelScape to real 3D PCD for two challenging 3D perception tasks. The promising results have shown that training DNNs on VoxelScape boosted the performance of the 3D perception tasks on the real PCD. Furthermore, we are also releasing the proposed data generation pipeline for the research community to facilitate realistic simulation of 3D LiDAR point cloud data for different scenarios beyond those covered in our VoxelScape dataset. The VoxelScape dataset and the corresponding LiDAR simulation codes are publicly available at https://voxel-scape.github.io/dataset

A weakly supervised inpainting-based learning method for lung CT image segmentation

Recently, various fully supervised learning methods are successfully applied for lung CT image segmentation. However, pixel-wise annotations are extremely expert-demanding and labor-intensive, but the performance of unsupervised learning methods are failed to meet the demands of practical applications. To achieve a reasonable trade-off between the performance and label dependency, a novel weakly supervised inpainting-based learning method is introduced, in which only bounding box labels are required for accurate segmentation. Specifically, lesion regions are first detected by an object detection network, then we crop them out of the input image and recover the missing holes to normal regions using a progressive CT inpainting network (PCIN). Finally, a post-processing method is designed to get the accurate segmentation mask from the difference image of input and recovered images. In addition, real information (i.e., number, location and size) of the bounding boxes of lesions from the dataset guides us to make the training dataset for PCIN. We apply a multi-scale supervised strategy to train PCIN for a progressive and stable inpainting. Moreover, to remove the visual artifacts resulted from the invalid features of missing holes, an initial patch generation network (IPGN) is proposed for holes initialization with generated pseudo healthy image patches. Experiments on the public COVID-19 dataset demonstrate that PCIN is outstanding in lung CT images inpainting, and the performance of our proposed weakly supervised method is comparable to fully supervised methods.

WeedNet-R: a sugar beet field weed detection algorithm based on enhanced RetinaNet and context semantic fusion.

Accurate and dependable weed detection technology is a prerequisite for weed control robots to do autonomous weeding. Due to the complexity of the farmland environment and the resemblance between crops and weeds, detecting weeds in the field under natural settings is a difficult task. Existing deep learning-based weed detection approaches often suffer from issues such as monotonous detection scene, lack of picture samples and location information for detected items, low detection accuracy, etc. as compared to conventional weed detection methods. To address these issues, WeedNet-R, a vision-based network for weed identification and localization in sugar beet fields, is proposed. WeedNet-R adds numerous context modules to RetinaNet's neck in order to combine context information from many feature maps and so expand the effective receptive fields of the entire network. During model training, meantime, a learning rate adjustment method combining an untuned exponential warmup schedule and cosine annealing technique is implemented. As a result, the suggested method for weed detection is more accurate without requiring a considerable increase in model parameters. The WeedNet-R was trained and assessed using the OD-SugarBeets dataset, which is enhanced by manually adding the bounding box labels based on the publicly available agricultural dataset, i.e. SugarBeet2016. Compared to the original RetinaNet, the mAP of the proposed WeedNet-R increased in the weed detection job in sugar beet fields by 4.65% to 92.30%. WeedNet-R's average precision for weed and sugar beet is 85.70% and 98.89%, respectively. WeedNet-R outperforms other sophisticated object detection algorithms in terms of detection accuracy while matching other single-stage detectors in terms of detection speed.

BBox-Guided Segmentor: Leveraging expert knowledge for accurate stroke lesion segmentation using weakly supervised bounding box prior.

Stroke is one of the leading causes of death and disability in the world. Despite intensive research on automatic stroke lesion segmentation from non-invasive imaging modalities including diffusion-weighted imaging (DWI), challenges remain such as a lack of sufficient labeled data for training deep learning models and failure in detecting small lesions. In this paper, we propose BBox-Guided Segmentor, a method that significantly improves the accuracy of stroke lesion segmentation by leveraging expert knowledge. Specifically, our model uses a very coarse bounding box label provided by the expert and then performs accurate segmentation automatically. The small overhead of having the expert provide a rough bounding box leads to large performance improvement in segmentation, which is paramount to accurate stroke diagnosis. To train our model, we employ a weakly-supervised approach that uses a large number of weakly-labeled images with only bounding boxes and a small number of fully labeled images. The scarce fully labeled images are used to train a generator segmentation network, while adversarial training is used to leverage the large number of weakly-labeled images to provide additional learning signals. We evaluate our method extensively using a unique clinical dataset of 99 fully labeled cases (i.e., with full segmentation map labels) and 831 weakly labeled cases (i.e., with only bounding box labels), and the results demonstrate the superior performance of our approach over state-of-the-art stroke lesion segmentation models. We also achieve competitive performance as a SOTA fully supervised method using less than one-tenth of the complete labels. Our proposed approach has the potential to improve stroke diagnosis and treatment planning, which may lead to better patient outcomes.

FedMix: Mixed Supervised Federated Learning for Medical Image Segmentation.

The purpose of federated learning is to enable multiple clients to jointly train a machine learning model without sharing data. However, the existing methods for training an image segmentation model have been based on an unrealistic assumption that the training set for each local client is annotated in a similar fashion and thus follows the same image supervision level. To relax this assumption, in this work, we propose a label-agnostic unified federated learning framework, named FedMix, for medical image segmentation based on mixed image labels. In FedMix, each client updates the federated model by integrating and effectively making use of all available labeled data ranging from strong pixel-level labels, weak bounding box labels, to weakest image-level class labels. Based on these local models, we further propose an adaptive weight assignment procedure across local clients, where each client learns an aggregation weight during the global model update. Compared to the existing methods, FedMix not only breaks through the constraint of a single level of image supervision but also can dynamically adjust the aggregation weight of each local client, achieving rich yet discriminative feature representations. Experimental results on multiple publicly-available datasets validate that the proposed FedMix outperforms the state-of-the-art methods by a large margin. In addition, we demonstrate through experiments that FedMix is extendable to multi-class medical image segmentation and much more feasible in clinical scenarios. The code is available at: https://github.com/Jwicaksana/FedMix.

Weakly and semi-supervised detection, segmentation and tracking of table grapes with limited and noisy data

Detection, segmentation and tracking of fruits and vegetables are three fundamental tasks for precision agriculture, enabling robotic harvesting and yield estimation applications. However, modern algorithms are data hungry and it is not always possible to gather enough data to apply the best performing supervised approaches. Since data collection is an expensive and cumbersome task, the enabling technologies for using computer vision in agriculture are often out of reach for small businesses. Following previous work in this context (Ciarfuglia et al., 2022), where we proposed an initial weakly supervised solution to reduce the data needed to get state-of-the-art detection and segmentation in precision agriculture applications, here we improve that system and explore the problem of tracking fruits in orchards. We present the case of vineyards of table grapes in southern Lazio (Italy) since grapes are a difficult fruit to segment due to occlusion, colour and general illumination conditions. We consider the case in which there is some initial labelled data that could work as source data (e.g. wine grape data), but it is considerably different from the target data (e.g. table grape data). To improve detection and segmentation on the target data, we propose to train the segmentation algorithm with a weak bounding box label, while for tracking we leverage 3D Structure from Motion algorithms to generate new labels from already labelled samples. Finally, the two systems are combined in a full semi-supervised approach. Comparisons with state-of-the-art supervised solutions show how our methods are able to train new models that achieve high performances with few labelled images and with very simple labelling.

Contrastive uncertainty based biomarkers detection in retinal optical coherence tomography images

Objective. Retinal biomarker in optical coherence tomography (OCT) images plays a key guiding role in the follow-up diagnosis and clinical treatment of eye diseases. Although there have been many deep learning methods to automatically process retinal biomarker, the detection of retinal biomarkers is still a great challenge due to the similar characteristics to normal tissue, large changes in size and shape and fuzzy boundary of different types of biomarkers. To overcome these challenges, a novel contrastive uncertainty network (CUNet) is proposed for retinal biomarkers detection in OCT images. Approach. In CUNet, proposal contrastive learning is designed to enhance the feature representation of retinal biomarkers, aiming at boosting the discrimination ability of network between different types of retinal biomarkers. Furthermore, we proposed bounding box uncertainty and combined it with the traditional bounding box regression, thereby improving the sensitivity of the network to the fuzzy boundaries of retinal biomarkers, and to obtain a better localization result. Main results. Comprehensive experiments are conducted to evaluate the performance of the proposed CUNet. The experimental results on two datasets show that our proposed method achieves good detection performance compared with other detection methods. Significance. We propose a method for retinal biomarker detection trained by bounding box labels. The proposal contrastive learning and bounding box uncertainty are used to improve the detection of retinal biomarkers. The method is designed to help reduce the amount of work doctors have to do to detect retinal diseases.

Labels are Not Perfect: Inferring Spatial Uncertainty in Object Detection

The availability of many real-world driving datasets is a key reason behind the recent progress of object detection algorithms in autonomous driving. However, there exist ambiguity or even failures in object labels due to error-prone annotation process or sensor observation noise. Current public object detection datasets only provide deterministic object labels without considering their inherent uncertainty, as does the common training process or evaluation metrics for object detectors. As a result, an in-depth evaluation among different object detection methods remains challenging, and the training process of object detectors is sub-optimal, especially in probabilistic object detection. In this work, we infer the uncertainty in bounding box labels from LiDAR point clouds based on a generative model, and define a new representation of the probabilistic bounding box through a spatial uncertainty distribution. Comprehensive experiments show that the proposed model reflects complex environmental noises in LiDAR perception and the label quality. Furthermore, we propose Jaccard IoU (JIoU) as a new evaluation metric that extends IoU by incorporating label uncertainty. We conduct an in-depth comparison among several LiDAR-based object detectors using the JIoU metric. Finally, we incorporate the proposed label uncertainty in a loss function to train a probabilistic object detector and to improve its detection accuracy. We verify our proposed methods on two public datasets (KITTI, Waymo), as well as on simulation data. Code is released at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/</uri> ZiningWang/Inferring-Spatial-Uncertainty-in-Object-Detection.

Learning-based three-dimensional registration with weak bounding box supervision.

Purpose: Image registration is the process of aligning images, and it is a fundamental task in medical image analysis. While many tasks in the field of image analysis, such as image segmentation, are handled almost entirely with deep learning and exceed the accuracy of conventional algorithms, currently available deformable image registration methods are often still conventional. Deep learning methods for medical image registration have recently reached the accuracy of conventional algorithms. However, they are often based on a weakly supervised learning scheme using multilabel image segmentations during training. The creation of such detailed annotations is very time-consuming. Approach: We propose a weakly supervised learning scheme for deformable image registration. By calculating the loss function based on only bounding box labels, we are able to train an image registration network for large displacement deformations without using densely labeled images. We evaluate our model on interpatient three-dimensional abdominal CT and MRI images. Results: The results show an improvement of (for CT images) and 20% (for MRI images) in comparison to the unsupervised method. When taking into account the reduced annotation effort, the performance also exceeds the performance of weakly supervised training using detailed image segmentations. Conclusion: We show that the performance of image registration methods can be enhanced with little annotation effort using our proposed method.

Bilateral Temporal Re-Aggregation for Weakly-Supervised Video Object Segmentation

Weakly-supervised video object segmentation is an emerging video task to track and segment the target given a simple bounding box label, which requires the method to fully catch and utilize the target information. Most existing approaches only rely on the guidance of a single frame and ignore the interaction between different frames when gathering information, making them hard to achieve reliable target representation. In this paper, we propose to capture the temporal dependencies and gather information from multiple frames through bilateral temporal re-aggregation. We explore three schemes to build the aggregation: 1) a two-stage re-aggregation mechanism is applied to provide target prior to the current frame, which obtains more valid feature matching and information aggregation; 2) a query-memory bilateral aggregation module is proposed to aggregate features from an unlimited amount of past frames and enable the mutual perception between different frames to validate the gathered information; 3) we guide the learning of aggregation modules through a novel cross-task representation distillation, transferring the knowledge from a semi-supervised model to our weakly-supervised model without increasing the inference latency. These schemes collaboratively build an efficient and competent aggregation process, thus we can fully exploit the video context to make the inference. Experimental results on four benchmarks show that our method achieves superior performance than previous methods and still maintains the efficiency ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$e.g$ </tex-math></inline-formula> ., overall scores of 70.4% and 72.5% on the YouTube-VOS and DAVIS 2017 validation sets, respectively).

Weakly-Supervised Salient Object Detection Using Point Supervision

Current state-of-the-art saliency detection models rely heavily on large datasets of accurate pixel-wise annotations, but manually labeling pixels is time-consuming and labor-intensive. There are some weakly supervised methods developed for alleviating the problem, such as image label, bounding box label, and scribble label, while point label still has not been explored in this field. In this paper, we propose a novel weakly-supervised salient object detection method using point supervision. To infer the saliency map, we first design an adaptive masked flood filling algorithm to generate pseudo labels. Then we develop a transformer-based point-supervised saliency detection model to produce the first round of saliency maps. However, due to the sparseness of the label, the weakly supervised model tends to degenerate into a general foreground detection model. To address this issue, we propose a Non-Salient Suppression (NSS) method to optimize the erroneous saliency maps generated in the first round and leverage them for the second round of training. Moreover, we build a new point-supervised dataset (P-DUTS) by relabeling the DUTS dataset. In P-DUTS, there is only one labeled point for each salient object. Comprehensive experiments on five largest benchmark datasets demonstrate our method outperforms the previous state-of-the-art methods trained with the stronger supervision and even surpass several fully supervised state-of-the-art models. The code is available at: https://github.com/shuyonggao/PSOD.

Hybrid supervised instance segmentation by learning label noise suppression

To reach top accuracy, current fully supervised instance segmentation methods severely rely on large-scale pixel-wise labeled datasets. They are usually expensive and time-consuming to obtain. Though weakly or semi-supervised methods utilize cheap bounding box labeled, image-level labeled or unlabeled samples to save the labeling cost, their performance is largely sacrificed. To save labeling cost without losing much performance, in this paper, we present a pipeline that can utilize economical bounding box labels and accurate pixel-wise labels in a hybrid way. Specifically, we design two ancillary models to learn label noise suppression and obtain accurate pseudo pixel-wise labels from bounding box labels for training. One is designed to suppress mislabeling between foreground and background, and the other is designed to suppress noise from mislabeling of instances. Moreover, we exploit category-aware spatial attention module, category constraint module, instance constraint module, and self-learning training approach to improve the accuracy of pseudo labels. Experiments on the PASCAL VOC 2012 and the Cityscapes datasets show that our method can achieve competitive performance with much less labeling cost.

SAR Ship Detection Dataset (SSDD): Official Release and Comprehensive Data Analysis

SAR Ship Detection Dataset (SSDD) is the first open dataset that is widely used to research state-of-the-art technology of ship detection from Synthetic Aperture Radar (SAR) imagery based on deep learning (DL). According to our investigation, up to 46.59% of the total 161 public reports confidently select SSDD to study DL-based SAR ship detection. Undoubtedly, this situation reveals the popularity and great influence of SSDD in the SAR remote sensing community. Nevertheless, the coarse annotations and ambiguous standards of use of its initial version both hinder fair methodological comparisons and effective academic exchanges. Additionally, its single-function horizontal-vertical rectangle bounding box (BBox) labels can no longer satisfy the current research needs of the rotatable bounding box (RBox) task and the pixel-level polygon segmentation task. Therefore, to address the above two dilemmas, in this review, advocated by the publisher of SSDD, we will make an official release of SSDD based on its initial version. SSDD’s official release version will cover three types: (1) a bounding box SSDD (BBox-SSDD), (2) a rotatable bounding box SSDD (RBox-SSDD), and (3) a polygon segmentation SSDD (PSeg-SSDD). We relabel ships in SSDD more carefully and finely, and then explicitly formulate some strict using standards, e.g., (1) the training-test division determination, (2) the inshore-offshore protocol, (3) the ship-size reasonable definition, (4) the determination of the densely distributed small ship samples, and (5) the determination of the densely parallel berthing at ports ship samples. These using standards are all formulated objectively based on the using differences of existing 75 (161 × 46.59%) public reports. They will be beneficial for fair method comparison and effective academic exchanges in the future. Most notably, we conduct a comprehensive data analysis on BBox-SSDD, RBox-SSDD, and PSeg-SSDD. Our analysis results can provide some valuable suggestions for possible future scholars to further elaborately design DL-based SAR ship detectors with higher accuracy and stronger robustness when using SSDD.

Query-Memory Re-Aggregation for Weakly-supervised Video Object Segmentation

Weakly-supervised video object segmentation (WVOS) is an emerging video task that can track and segment the target given a simple bounding box label. However, existing WVOS methods are still unsatisfied in either speed or accuracy, since they only use the exemplar frame to guide the prediction while they neglect the reference from other frames. To solve the problem, we propose a novel Re-Aggregation based framework, which uses feature matching to efficiently find the target and capture the temporal dependencies from multiple frames to guide the segmentation. Based on a two-stage structure, our framework builds an information-symmetric matching process to achieve robust aggregation. In each stage, we design a Query-Memory Aggregation (QMA) module to gather features from the past frames and make bidirectional aggregation to adaptively weight the aggregated features, which relieves the latent misguidance in unidirectional aggregation. To further exploit the information from different aggregation stages, we propose a novel coarse-fine constraint by using the Cascaded Refinement Module (CRM) to combine the predictions from different stages and further boosts the performance. Experimental results on three benchmarks show that our method achieves the state-of-the-art performance in WVOS (e.g., an overall score of 84.7% on the DAVIS 2016 validation set).