Connected Conditional Random Field Research Articles

Relationship detection for manipulation in object stacking scene with fully connected CRF

Relationship detection for manipulation in object stacking scenes is an actively studied problem in recent years. Since the stacked objects may occlude and block each other, it is necessary to model relationships among them for safety. Almost all existing solutions to this problem model object relationships separately and ignore their dependencies on each other. In this paper, we propose a fully connected Conditional Random Field (CRF) to impose global constraints on the object stacking scenes. Since the inference of fully connected CRF is usually intractable, we propose efficient methods for the exact inference and the variational inference to solve the proposed CRFs. Furthermore, the variational inference is implemented as an RNN for end-to-end training. Experimental results on Visual Manipulation Relationship Dataset VMRD and REGRAD suggest that the proposed CRF can, with a clear margin, improve the accuracy of manipulation relationship detection in object stacking scenes. Finally, the proposed models are also deployed on a physical robot to build a relation-aware grasping system. Results demonstrate that our algorithms can be generalized to real-world scenes well and help the robot perform safe grasping in object stacking scenes.

Segmentation of wheat scab fungus spores based on CRF_ResUNet++

Accurate segmentation of tiny spore pictures is a significant technique in computer-aided evaluation of wheat scab. A CRF_ResUNet++ spore segmentation model using UNet++ as the fundamental framework is proposed to address the issues of uneven brightness, low spore-background contrast, and spore adhesion in microscopically collected Fusarium spore images. ResNet and fully connected conditional random fields (CRF) are combined in this model to address these issues. First, a residual block ResNet is introduced into UNet++ to enhance the propagation of features and extract more spore detail information. Next, the spore image is initially segmented using UNet++ with an encoder-decoder structure. Finally, postprocessing is carried out using a fully connected conditional random field model to obtain more precise edges and complete spore regions. The laboratory collection dataset test results revealed that the F1-score and MIoU reached 0.943 and 0.925, respectively, with an average detection Precision improvement of 3.2 % over the original UNet++ model. In comparison to existing models, the suggested model can segment spore pictures in challenging circumstances, such as when they contain independent and inter-adherent spores, with excellent computer vision processing capabilities effects and objective evaluation metric segmentation effects. In conclusion, the method in this study can reliably detect and segment adherent wheat scab spores in complex backgrounds, which offers critical technical support for automatic detection of adherent wheat scab spores and early spore concentration prediction in complex environments in the field.

A New Image Segmentation Method Based on the YOLO5 and Fully Connected CRF

When manually polishing blades, skilled workers can quickly machine a blade by observing the characteristics of the polishing sparks. To help workers better recognize spark images, we used an industrial charge-coupled device (CCD) camera to capture the spark images. Firstly, the spark image region detected by yolo5, then segment from the background. Secondly, the target region was further segmented and refined in a fully connected conditional random field (CRF), from which the complete spark image obtained. Experimental results showed that this method could quickly and accurately segment whole spark image. The test results showed that this method was better than other image segmentation algorithms. Our method could better segment irregular image, improve recognition and segmentation efficiency of spark image, achieve automatic image segmentation, and replace human observation.

LeNo: Adversarial Robust Salient Object Detection Networks with Learnable Noise

Pixel-wise prediction with deep neural network has become an effective paradigm for salient object detection (SOD) and achieved remarkable performance. However, very few SOD models are robust against adversarial attacks which are visually imperceptible for human visual attention. The previous work robust saliency (ROSA) shuffles the pre-segmented superpixels and then refines the coarse saliency map by the densely connected conditional random field (CRF). Different from ROSA that rely on various pre- and post-processings, this paper proposes a light-weight Learnable Noise (LeNo) to defend adversarial attacks for SOD models. LeNo preserves accuracy of SOD models on both adversarial and clean images, as well as inference speed. In general, LeNo consists of a simple shallow noise and noise estimation that embedded in the encoder and decoder of arbitrary SOD networks respectively. Inspired by the center prior of human visual attention mechanism, we initialize the shallow noise with a cross-shaped gaussian distribution for better defense against adversarial attacks. Instead of adding additional network components for post-processing, the proposed noise estimation modifies only one channel of the decoder. With the deeply-supervised noise-decoupled training on state-of-the-art RGB and RGB-D SOD networks, LeNo outperforms previous works not only on adversarial images but also on clean images, which contributes stronger robustness for SOD. Our code is available at https://github.com/ssecv/LeNo.

Automatic detection and recognition of nasopharynx gross tumour volume (GTVnx) by deep learning for nasopharyngeal cancer radiotherapy through magnetic resonance imaging

BackgroundIn this study, we propose the deep learning model-based framework to automatically delineate nasopharynx gross tumor volume (GTVnx) in MRI images.MethodsMRI images from 200 patients were collected for training-validation and testing set. Three popular deep learning models (FCN, U-Net, Deeplabv3) are proposed to automatically delineate GTVnx. FCN was the first and simplest fully convolutional model. U-Net was proposed specifically for medical image segmentation. In Deeplabv3, the proposed Atrous Spatial Pyramid Pooling (ASPP) block, and fully connected Conditional Random Field(CRF) may improve the detection of the small scattered distributed tumor parts due to its different scale of spatial pyramid layers. The three models are compared under same fair criteria, except the learning rate set for the U-Net. Two widely applied evaluation standards, mIoU and mPA, are employed for the detection result evaluation.ResultsThe extensive experiments show that the results of FCN and Deeplabv3 are promising as the benchmark of automatic nasopharyngeal cancer detection. Deeplabv3 performs best with the detection of mIoU 0.8529 ± 0.0017 and mPA 0.9103 ± 0.0039. FCN performs slightly worse in term of detection accuracy. However, both consume similar GPU memory and training time. U-Net performs obviously worst in both detection accuracy and memory consumption. Thus U-Net is not suggested for automatic GTVnx delineation.ConclusionsThe proposed framework for automatic target delineation of GTVnx in nasopharynx bring us the desirable and promising results, which could not only be labor-saving, but also make the contour evaluation more objective. This preliminary results provide us with clear directions for further study.

Meta-Knowledge Guided Weakly Supervised Instance Segmentation for Optical and SAR Image Interpretation

The interpretation of optical and synthetic aperture radar (SAR) images in remote sensing is general for many tasks, such as environmental monitoring, marine management, and resource planning. Instance segmentation of optical and SAR images, which can simultaneously provide instance-level localization and pixel-level classification of objects of interest, is a crucial and challenging task in image interpretation. Considering that most current methods for instance segmentation of optical and SAR images rely on expensive pixel-level annotation, we develop a weakly supervised instance segmentation (WSIS) method to balance the visual processing requirements with the annotation cost. First, we decompose the prior knowledge of the mask-aware task in WSIS into three meta-knowledge components: fundamental knowledge, apparent knowledge, and detailed knowledge inspired by human visual perception habits of “whole to part” and “coarse to detailed.” Then, a meta-knowledge-guided weakly supervised instance segmentation network (MGWI-Net) is proposed. In this network, the weakly supervised mask (WSM) head can instantiate both fundamental knowledge and apparent knowledge to perform mask awareness without any annotations at the pixel level. The network also includes a mask information awareness assist (MIAA) head, which can implicitly guide the network to learn detailed information about edges through the boundary-sensitive feature of the fully connected conditional random field (CRF), facilitating the instantiation of detailed knowledge. The experimental results show that the MGWI-Net can efficiently generate instance masks for optical and SAR images and achieve the approximate instance segmentation results of the fully supervised method with about one-eighth of the annotation production time. The model parameters and processing speed of our network are also competitive. This study can provide inexpensive and convenient technical support for applying and promoting instance segmentation methods for optical and SAR images.

Water–Land Segmentation via Structure-Aware CNN–Transformer Network on Large-Scale SAR Data

Water–land segmentation has important applications in many aspects, such as environmental protection and flood prevention. Synthetic aperture radar (SAR) is one of the important tools for watershed monitoring. However, neither the traditional SAR water–land segmentation methods nor convolutional neural network (CNN)-based algorithms have satisfactory accuracy for the boundaries of water, especially for fine rivers/lakes. Transformer can effectively extract global information and has achieved amazing success in image interpretation. In this article, a structure-aware CNN–Transformer network with U-shaped structure is designed to solve the water segmentation problem, named Trans-SANet. Also, an edge refinement postprocessing module is added to Trans-SANet to refine the water edges, named Trans-SANet+. First, a cascaded attentional upsampler (CAUP) is designed in the decoder to combine spatial and temporal attentional serialization to fully capture feature information and avoid feature omission. Second, a structure-aware loss (SA-Loss) function is designed to give sufficient attention to the overall boundary structure while avoiding the loss of small-scale features to reduce the segmentation error of the boundary, thus improving the segmentation accuracy of the fine water region. Finally, combining fully connected conditional random field (CRF) with edge extraction, a postprocessing module is proposed to further optimize the results of water–land segmentation. Experiments on three SAR datasets show that Trans-SANet+ outperforms several most advanced depth segmentation networks in terms of accuracy and speed.

A TWO-STAGE EXTRACTION NETWORK FOR THE OPTICAL COHERENCE TOMOGRAPHY CONTOUR

Optical coherence tomography (OCT) with high resolution was found lucrative for coronary heart disease diagnosis and treatment such as evaluating the degree of vascular stenosis and guiding stent implantation. Lumen contour is an essential information for assessing the shape and stenosis of blood vessels. The automatic and accurate extraction of lumen contour can significantly reduce the burden of doctors and improve the efficiency of diagnosis and treatment of coronary heart disease. The existing contour segmentation methods have poor performance in extracting the contour of the image with a missing blood vessel wall. Therefore, this paper proposes a two-stage contour extraction method based on deep learning. First, the method uses the basic contour extraction network to complete the feature image extraction and coarse contour segmentation, then inputs the feature image into the attentional class feature (ACF) module, and uses the coarse segmentation results to guide the generation of the new feature map for the refined contour segmentation. Finally, the fully connected conditional random field (CRF) module is used for post-processing yielding a more refined and smooth segmentation boundary. The experiments show that compared with the other methods, the lumen contour segmentation method proposed in this paper can adequately resolve the problem of partial vessel wall absence in OCT images and improve the accuracy of lumen contour extraction.

PREFACE: A SPECIAL SELECTION ON EMERGING TECHNIQUES FOR BIOMECHANICS–PART II

PREFACE: A SPECIAL SELECTION ON EMERGING TECHNIQUES FOR BIOMECHANICS–PART II

Efficient two-step liver and tumour segmentation on abdominal CT via deep learning and a conditional random field

Segmentation of the liver and tumours from computed tomography (CT) scans is an important task in hepatic surgical planning. Manual segmentation of the liver and tumours is a time-consuming and labour-intensive task; therefore, a fully automated method for performing this segmentation is particularly desired. An automatic two-step liver and tumour segmentation method is presented in this paper. A cascade framework is used during the segmentation process, and a fully connected conditional random field (CRF) method is used to refine the tumour segmentation result. First, the proposed fractal residual U-Net (FRA-UNet) is used to locate and initially segment the liver. Then, FRA-UNet is further used to predict liver tumours from the liver region of interest (ROI). Finally, a three-dimensional (3D) CRF is used to refine the tumour segmentation results. The improved fractal residual (FR) structure effectively retains more effective features for improving the segmentation performance of deeper networks, the improved deep residual block can utilise the feature information more effectively, and the 3D CRF method smooths the contours and avoids the tumour oversegmentation problem. FRA-UNet is tested on the Liver Tumour Segmentation challenge dataset (LiTS) and the 3D Image Reconstruction for Comparison of Algorithm Database dataset (3DIRCADb), achieving 97.13% and 97.18% Dice similarity coefficients (DSCs) for liver segmentation and 71.78% and 68.97% DSCs for tumour segmentation, respectively, outperforming most state-of-the-art networks.

A new frog leaping algorithm-oriented fully convolutional neural network for dance motion object saliency detection

Image saliency detection is an important research topic in the field of computer vision. With the traditional saliency detection models, the texture details are not obvious and the edge contour is not complete. The accuracy and recall rate of object detection are low, which are mostly based on the manual features and prior information. With the rise of deep convolutional neural networks, saliency detection has been rapidly developed. However, the existing saliency methods still have some common shortcomings, and it is difficult to uniformly highlight the clear boundary and internal region of the whole object in complex images, mainly because of the lack of sufficient and rich features. In this paper, a new frog leaping algorithm-oriented fully convolutional neural network is proposed for dance motion object saliency detection. The VGG (Visual Geometry Group) model is improved. The final full connection layer is removed, and the jump connection layer is used for the saliency prediction, which can effectively combine the multi-scale information from different convolution layers in the convolutional neural network. Meanwhile, an improved frog leaping algorithm is used to optimize the selection of initial weights during network initialization. In the process of network iteration, the forward propagation loss of convolutional neural network is calculated, and the anomaly weight is corrected by using the improved frog leaping algorithm. When the network satisfies the terminal conditions, the final weight is optimized by one frog leaping to make the network weight further optimization. In addition, the new network can combine high-level semantic information and low-level detail information in a data-driven framework. In order to preserve the unity of the object boundary and inner region effectively, the fully connected conditional random field (CRF) model is used to adjust the obtained saliency feature map. In this paper, the precision recall (PR) curve, F-measure, maximum F-measure, weighted F-measure and mean absolute error (MAE) are tested on six widely used public data sets. Compared with other most advanced and representative methods, the results show that the proposed method achieves better performance and it is superior to most representative methods. The presented method reveals that it has strong robustness for image saliency detection with various scenes, and can make the boundary and inner region of the saliency object more uniform and the detection results more accurate.

A Correlation Context-Driven Method for Sea Fog Detection in Meteorological Satellite Imagery

Sea fog detection is a challenging and essential issue in satellite remote sensing. Although conventional threshold methods and deep learning methods can achieve pixel-level classification, it is difficult to distinguish ambiguous boundaries and thin structures from the background. Considering the correlations between neighbor pixels and the affinities between superpixels, a correlation context-driven method for sea fog detection is proposed in this letter, which mainly consists of a two-stage superpixel-based fully convolutional network (SFCNet), named SFCNet. A fully connected Conditional Random Field (CRF) is utilized to model the dependencies between pixels. To alleviate the problem of high cloud occlusion, an attentive Generative Adversarial Network (GAN) is implemented for image enhancement by exploiting contextual information. Experimental results demonstrate that our proposed method achieves 91.65% mIoU and obtains more refined segmentation results, performing well in detecting fogs in small, broken bits and weak contrast thin structures, as well as detects more obscured parts.

A Semisupervised CRF Model for CNN-Based Semantic Segmentation With Sparse Ground Truth

Convolutional neural networks (CNNs) represent the new reference approach for semantic segmentation of very-high-resolution (VHR) images, due to their ability to automatically capture semantic information while learning relevant features. However, as for most supervised methods, the map accuracy depends on the quantity and quality of ground truth (GT) used to train them. The use of densely annotated data (i.e., a detailed, exhaustive, pixel-level GT) allows to obtain effective CNN models but normally implies high efforts in annotation. Such ground truth is often available in benchmark datasets on which new methods are tested, but not on real data for land-cover applications, where only sparse annotations might be sufficiently cost effective. A CNN model trained with such incomplete GT maps has the tendency to smooth object boundaries because they are never precisely delineated in the GT. To cope with those shortcomings, we propose to exploit the intermediate activation maps of the CNN and to deploy a semisupervised fully connected conditional random field (CRF). In comparison with competitors using the same sparse annotations, the proposed method is able to better fill part of the performance gap compared to a CNN trained on the densely annotated, but generally unavailable, GTs.

3D convolution for multidate crop recognition from multitemporal image sequences

ABSTRACT The increasing food demand is regarded as the main threat to nature today. In this scenario, Remote Rensing is an essential technology to assess and monitor the extent and productivity of cultivated land. However, most work on crop mapping from remote sensing imagery so far focused on temperate areas, where a single crop prevails throughout the year. This paper explores different Convolutional Neural Network architectures for multi-temporal crop recognition in tropical areas, with highly complex crop dynamics and many crops per year. The proposed models are a U-Net with bidirectional ConvLSTM added to its skip connections, a U-Net with 3D convolutions, and a U-Net with 3D convolutions and Atrous Temporal Pyramid Pooling. The paper also assesses the gain of a post-processing step based on fully connected conditional random fields (CRF) as a second look at the spatial context to improve the networks’ predictions. In an extensive experimental analysis conducted on two public multitemporal datasets of tropical areas with complex crop dynamics, all proposed models achieved better results than state-of-the-art methods based on recurrent networks in terms of Overall Accuracy and Average F1-Score. The 3D convolution with the Atrous Temporal Pyramid Pooling model stood up as the best performing architecture, with gains up to 4.3% over the baseline. The proposed CRF post-processing also proved beneficial for all tested architectures, bringing improvements of up to 2.7% in terms of average F1-score.

Paddy Rice Mapping in Thailand Using Time-Series Sentinel-1 Data and Deep Learning Model

The elimination of hunger is the top concern for developing countries and is the key to maintain national stability and security. Paddy rice occupies an essential status in food supply, whose accurate monitoring is of great importance for human sustainable development. As one of the most important paddy rice production countries in the world, Thailand has a favorable hot and humid climate for paddy rice growing, but the growth patterns of paddy rice are too complicated to construct promising growth models for paddy rice discrimination. To solve this problem, this study proposes a large-scale paddy rice mapping scheme, which uses time-series Sentinel-1 data to generate a convincing annual paddy rice map of Thailand. The proposed method extracts temporal statistical features of the time-series SAR images to overcome the intra-class variability due to different management practices and modifies the U-Net model with the fully connected Conditional Random Field (CRF) to maintain the edge of the fields. In this study, 758 Sentinel-1 images that covered the whole country from the end of 2018 to 2019 were acquired to generate the annual paddy rice map. The accuracy, precision, and recall of the resultant paddy rice map reached 91%, 87%, and 95%, respectively. Compared to SVM classifier and the U-Net model based on feature selection strategy (FS-U-Net), the proposed scheme achieved the best overall performance, which demonstrated the capability of overcoming the complex cultivation conditions and accurately identifying the fragmented paddy rice fields in Thailand. This study provides a promising tool for large-scale paddy rice monitoring in tropical production regions and has great potential in the global sustainable development of food and environment management.

A Stacking Ensemble Deep Learning Model for Building Extraction from Remote Sensing Images

Automatically extracting buildings from remote sensing images with deep learning is of great significance to urban planning, disaster prevention, change detection, and other applications. Various deep learning models have been proposed to extract building information, showing both strengths and weaknesses in capturing the complex spectral and spatial characteristics of buildings in remote sensing images. To integrate the strengths of individual models and obtain fine-scale spatial and spectral building information, this study proposed a stacking ensemble deep learning model. First, an optimization method for the prediction results of the basic model is proposed based on fully connected conditional random fields (CRFs). On this basis, a stacking ensemble model (SENet) based on a sparse autoencoder integrating U-NET, SegNet, and FCN-8s models is proposed to combine the features of the optimized basic model prediction results. Utilizing several cities in Hebei Province, China as a case study, a building dataset containing attribute labels is established to assess the performance of the proposed model. The proposed SENet is compared with three individual models (U-NET, SegNet and FCN-8s), and the results show that the accuracy of SENet is 0.954, approximately 6.7%, 6.1%, and 9.8% higher than U-NET, SegNet, and FCN-8s models, respectively. The identification of building features, including colors, sizes, shapes, and shadows, is also evaluated, showing that the accuracy, recall, F1 score, and intersection over union (IoU) of the SENet model are higher than those of the three individual models. This suggests that the proposed ensemble model can effectively depict the different features of buildings and provides an alternative approach to building extraction with higher accuracy.

An efficient two-step multi-organ registration on abdominal CT via deep-learning based segmentation

Registration of three dimensional (3D) abdominal computed tomography (CT) scans is essential for computer-aided disease diagnosis and treatment, but the non-rigid respiratory movements of abdomen increase its difficulty. An automatic two-step multi-organ registration method is presented in this paper for abdominal CT scans. First, the lightweight squeeze-and-excitation (SE) attention blocks and the fully connected conditional random field (CRF)-based post-processing are integrated into a fully convolutional networks (FCN) based model, which can achieve more accurate segmentation results for abdominal multiple organs such as liver, kidneys, and spleen. Then, a non-rigid local correlation coefficient (LCC) similarity metric and an isotropic total variation regularization are combined to register the multi-organ regions, which can reduce computation time and avoid the over-smooth problem of deformation field. The proposed method is validated on three public abdominal CT databases, and the experimental results show that the segmentation and registration performances of our method outperform those of some competing methods. Moreover, compared with global registration strategy, the average registration time of the proposed method is shortened by 20.45%, and the average MSE, PSNR, SSIM, and DICE values are also improved by 7.21%, 22.37%, 15.80%, and 14.43%, respectively.

Automated segmentation of macular edema for the diagnosis of ocular disease using deep learning method

Macular edema is considered as a major cause of visual loss and blindness in patients with ocular fundus diseases. Optical coherence tomography (OCT) is a non-invasive imaging technique, which has been widely applied for diagnosing macular edema due to its non-invasive and high resolution properties. However, the practical applications remain challenges due to the distorted retinal morphology and blurred boundaries near macular edema. Herein, we developed a novel deep learning model for the segmentation of macular edema in OCT images based on DeepLab framework (OCT-DeepLab). In this model, we used atrous spatial pyramid pooling (ASPP) to detect macular edema at multiple features and used the fully connected conditional random field (CRF) to refine the boundary of macular edema. OCT-DeepLab model was compared against the traditional hand-crafted methods (C-V and SBG) and the end-to-end methods (FCN, PSPnet, and U-net) to estimate the segmentation performance. OCT-DeepLab showed great advantage over the hand-crafted methods (C-V and SBG) and end-to-end methods (FCN, PSPnet, and U-net) as shown by higher precision, sensitivity, specificity, and F1-score. The segmentation performance of OCT-DeepLab was comparable to that of manual label, with an average area under the curve (AUC) of 0.963, which was superior to other end-to-end methods (FCN, PSPnet, and U-net). Collectively, OCT-DeepLab model is suitable for the segmentation of macular edema and assist ophthalmologists in the management of ocular disease.

WTS: A Weakly towards Strongly Supervised Learning Framework for Remote Sensing Land Cover Classification Using Segmentation Models

Land cover classification is one of the most fundamental tasks in the field of remote sensing. In recent years, fully supervised fully convolutional network (FCN)-based semantic segmentation models have achieved state-of-the-art performance in the semantic segmentation task. However, creating pixel-level annotations is prohibitively expensive and laborious, especially when dealing with remote sensing images. Weakly supervised learning methods from weakly labeled annotations can overcome this difficulty to some extent and achieve impressive segmentation results, but results are limited in accuracy. Inspired by point supervision and the traditional segmentation method of seeded region growing (SRG) algorithm, a weakly towards strongly (WTS) supervised learning framework is proposed in this study for remote sensing land cover classification to handle the absence of well-labeled and abundant pixel-level annotations when using segmentation models. In this framework, only several points with true class labels are required as the training set, which are much less expensive to acquire compared with pixel-level annotations through field survey or visual interpretation using high-resolution images. Firstly, they are used to train a Support Vector Machine (SVM) classifier. Once fully trained, the SVM is used to generate the initial seeded pixel-level training set, in which only the pixels with high confidence are assigned with class labels whereas others are unlabeled. They are used to weakly train the segmentation model. Then, the seeded region growing module and fully connected Conditional Random Fields (CRFs) are used to iteratively update the seeded pixel-level training set for progressively increasing pixel-level supervision of the segmentation model. Sentinel-2 remote sensing images are used to validate the proposed framework, and SVM is selected for comparison. In addition, FROM-GLC10 global land cover map is used as training reference to directly train the segmentation model. Experimental results show that the proposed framework outperforms other methods and can be highly recommended for land cover classification tasks when the pixel-level labeled datasets are insufficient by using segmentation models.

On the Effectiveness of Weakly Supervised Semantic Segmentation for Building Extraction From High-Resolution Remote Sensing Imagery

A critical obstacle to achieve semantic segmentation of remote sensing images by the deep convolutional neural network is the requirement of huge pixel-level labels. Taking building extraction as an example, this study focuses on how to effectively apply weakly supervised semantic segmentation (WSSS) to high-resolution remote sensing (HR) images with image-level labels, which is a prominent solution for the huge labeling challenge. The widely used two-step WSSS framework is adopted, in which the pseudo-masks are first produced from image-level labels and followed by a segmentation network trained by the pseudo-masks. In addition, the fully connected conditional random field (CRF) is utilized to explore spatial context in both training and prediction stages. Detailed analyzes are implemented on applying WSSS on HR images in terms of producing pseudo-masks, training segmentation network, and optimizing predictions. We show that the tradeoff between precision and recall of pseudo-masks, as well as the boundary accuracy and the background, needs to be carefully considered. The benefits of the segmentation network in the two-step framework are demonstrated in comparison to using classification network only for WSSS, and the effects of CRF-loss are identified to be powerful for improving the segmentation network while it is not appropriate for dense buildings. An overlapping strategy and CRF postprocessing are further demonstrated to be effective for optimizing the segmentation results during inferencing. Through deliberate settings, we can generate results comparable to fully supervised on the ISPRS Potsdam and Vaihingen dataset, which is meaningful for promoting WSSS applications for extracting geographic information from HR images.