Depth Features Research Articles

Gradient residual attention network for infrared image super-resolution

Image super-resolution techniques overcome the limitations of physical limits on infrared imaging systems and allow higher-resolution images of targets to be obtained on top of existing systems. In this paper, we proved that infrared images are more reconstructed than visible images based on data compression. We propose a CNN-based Gradient Residual Attention Network (GRAN) for infrared image super-resolution. Specifically, the residual dense module (RDB) is used to acquire depth features and the gradient operator (GO) gains fine-grained detail features. Meanwhile, the 3D attention block (3DAB) learns features' channel and spatial correlation to selectively capture more useful informative features. The experimental results show that the proposed method performs well compared to popular image super-resolution methods in terms of quantitative metrics and visual quality.

EFDCNet: Encoding fusion and decoding correction network for RGB-D indoor semantic segmentation

Semantic segmentation is a crucial task in vision measurement systems that involves understanding and segmenting different objects and regions within an image. Over the years, numerous RGB-D semantic segmentation methods have been developed, leveraging the encoder-decoder architecture to achieve outstanding performance. However, existing methods have two main problems that constrain further performance improvement. Firstly, in the encoding stage, existing methods have a weak ability to fuse cross-modal information, and low-quality depth maps can easily lead to poor feature representation. Secondly, in the decoding stage, the upsampling of high-level semantic information may cause the loss of contextual information, and low-level features from the encoder may bring noises to the decoder through skip connections. To solve these issues, we propose a novel Encoding Fusion and Decoding Correction Network (EFDCNet) for RGB-D indoor semantic segmentation. First, in the encoding stage of EFDCNet, we focus on extracting valuable information from low-quality depth maps, and employ a channel-wise filter to select informative depth features. Additionally, we establish the global dependencies between RGB and depth features via the self-attention mechanism to enhance the cross-modal feature interactions, extracting discriminant and powerful features. Then, in the decoding stage of EFDCNet, we use the highest-level information as semantic guidance to compensate for the upsampling information and filter out noise from the low-level encoder features propagated through the skip connections to the decoder. Extensive experiments conducted on two widely-used RGB-D indoor semantic segmentation datasets demonstrate that the proposed EFDCNet surpasses the performance of relevant state-of-the-art methods. The code is available at https://github.com/ Mark9010/EFDCNet

Depth-Quality Purification Feature Processing for Red Green Blue-Depth Salient Object Detection

With the advances in deep learning technology, Red Green Blue-Depth (RGB-D) Salient Object Detection (SOD) based on convolutional neural networks (CNNs) is gaining more and more attention. However, the accuracy of current models is challenging. It has been found that the quality of the depth features profoundly affects the accuracy. Several current RGB-D SOD techniques do not consider the quality of the depth features and directly fuse the original depth features and Red Green Blue (RGB) features for training, resulting in enhanced precision of the model. To address this issue, we propose a depth-quality purification feature processing network for RGB-D SOD, named DQPFPNet. First, we design a depth-quality purification feature processing (DQPFP) module to filter the depth features in a multi-scale manner and fuse them with RGB features in a multi-scale manner. This module can control and enhance the depth features explicitly in the process of cross-modal fusion, avoiding injecting noise or misleading depth features. Second, to prevent overfitting and avoid neuron inactivation, we utilize the RReLU activation function in the training process. In addition, we introduce the pixel position adaptive importance (PPAI) loss, which integrates local structure information to assign different weights to each pixel, thus better guiding the network’s learning process and producing clearer details. Finally, a dual-stage decoder is designed to utilize contextual information to improve the modeling ability of the model and enhance the efficiency of the network. Extensive experiments on six RGB-D datasets demonstrate that DQPFPNet outperforms recent efficient models and delivers cutting-edge accuracy.

Vision-Based Object Localization and Classification for Electric Vehicle Driving Assistance

The continuous advances in intelligent systems and cutting-edge technology have greatly influenced the development of intelligent vehicles. Recently, integrating multiple sensors in cars has improved and spread the advanced drive-assistance systems (ADAS) solutions for achieving the goal of total autonomy. Despite current self-driving approaches and systems, autonomous driving is still an open research issue that must guarantee the safety and reliability of drivers. This work employs images from two cameras and Global Positioning System (GPS) data to propose a 3D vision-based object localization and classification method for assisting a car during driving. The experimental platform is a prototype of a two-sitter electric vehicle designed and assembled for navigating the campus under controlled mobility conditions. Simultaneously, color and depth images from the primary camera are combined to extract 2D features, which are reprojected into 3D space. Road detection and depth features isolate point clouds representing the objects to construct the occupancy map of the environment. A convolutional neural network was trained to classify typical urban objects in the color images. Experimental tests validate car and object pose in the occupancy map for different scenarios, reinforcing the car position visually estimated with GPS measurements.

An improved YOLOv7 network using RGB-D multi-modal feature fusion for tea shoots detection

Due to the increasing scarcity of tea pickers, the implementation of intelligent harvesting for premium tea is a crucial prerequisite for the sustainable development of the premium tea industry. The initial step towards achieving intelligent and precise harvesting is the accurate detection of tender shoots, which consist of one bud and one leaf. However, accurately identifying tea shoots poses a challenging visual task due to their small size, variable shapes, as well as similar colors and backgrounds. The existing model, based on RGB images, can only detect partial targets. To address this issue and further enhance the detection of tea buds, this study proposes the utilization of multi-modal features encompassing red, green, blue, and depth (RGB-D) for identification. In addition, a unidirectional complementary multi-modal fusion method is introduced to minimize the adverse effects caused by low-quality depth information. Firstly, an RGB-D dataset comprising high-quality tea leaves is constructed, and the samples are carefully calibrated. Subsequently, an enhanced end-to-end RGB-D multi-modal object detection network, referred to as YOLO-RGBDtea, is developed based on You Only Look Once version 7 (YOLOv7). This model incorporates a parallel lightweight depth image feature extraction backbone network and incorporates a self-attention mechanism to prioritize contextual information. Lastly, a cross-modal spatial attention fusion module (CSFM) is devised to collaboratively integrate depth features with RGB features in a unidirectional manner. The experimental results reveal that YOLO-RGBDtea achieves an AP50 of 91.12% when confronting complex outdoor tea shoots, exhibiting significant performance improvements compared to YOLOv7, especially in scenarios involving small targets, overlapping target groups, and highly overexposed images. Notably, the parameter increment in YOLO-RGBDtea compared to the original YOLOv7 model is merely 17.8%, and the additional components can be seamlessly transferred to other models. Overall, this study introduces a straightforward yet effective multi-modal fusion method that bears theoretical and practical significance in advancing the detection of high-quality tea shoots in complex outdoor environments.

SiamFCA: A new fish single object tracking method based on siamese network with coordinate attention in aquaculture

In aquaculture, tracking fish helps to obtain information about their growth, behavior, location and health, etc. However, there are some problems in the process of fish tracking. Firstly, to solve the noise problem of underwater images, we use the contrast limited adaptive histogram equalization (CLAHE) to carry out local histogram equalization in each small block, which increases the image contrast while retaining more details. This operation enhances the contrast of the image while suppressing noise to solve the problem of noisy images. Then, to solve the problems of similarity interference, occlusion and scale variation among fish, we propose a new method for fish single object tracking in aquaculture environments, called SiamFCA, by combining the coordinated attention mechanism (CA). The method is based on the siamese network framework and uses a modified AlexNet to extract the depth features of image pairs. And through the coordinate attention mechanism, the model can learn the relationship between different objects more accurately, which can improve the accuracy and robustness of the model when dealing with complex image scenes. Finally, the correlation operation between classification and regression is performed based on the region proposal network (RPN). By classifying, region suggestions can be labeled as positive or negative samples, and regression can calibrate the position and size of the region proposal to match targets more accurately. The experimental results show that the precision and success rate of the algorithm are 84.2 % and 62.4 %, respectively, which are better than the tracking methods such as SiamRPN and DaSiamRPN, etc. The SiamFCA runs at 83 fps, which meets the demand of real-time fish tracking in aquaculture.

Depth-aware pose estimation using deep learning for exoskeleton gait analysis

In rehabilitation medicine, real-time analysis of the gait for human wearing lower-limb exoskeleton rehabilitation robot during walking can effectively prevent patients from experiencing excessive and asymmetric gait during rehabilitation training, thereby avoiding falls or even secondary injuries. To address the above situation, we propose a gait detection method based on computer vision for the real-time monitoring of gait during human–machine integrated walking. Specifically, we design a neural network model called GaitPoseNet, which is used for posture recognition in human–machine integrated walking. Using RGB images as input and depth features as output, regression of joint coordinates through depth estimation of implicit supervised networks. In addition, joint guidance strategy (JGS) is designed in the network framework. The degree of correlation between the various joints of the human body is used as a detection target to effectively overcome prediction difficulties due to partial joint occlusion during walking. Finally, a post processing algorithm is designed to describe patients’ walking motion by combining the pixel coordinates of each joint point and leg length. Our advantage is that we provide a non-contact measurement method with strong universality, and use depth estimation and JGS to improve measurement accuracy. Conducting experiments on the Walking Pose with Exoskeleton (WPE) Dataset shows that our method can reach 95.77% PCKs@0.1, 93.14% PCKs@0.08 and 3.55 ms runtime. Therefore our method achieves advanced performance considering both speed and accuracy.

Highway Visibility Estimation in Foggy Weather via Multi-Scale Fusion Network.

Poor visibility has a significant impact on road safety and can even lead to traffic accidents. The traditional means of visibility monitoring no longer meet the current needs in terms of temporal and spatial accuracy. In this work, we propose a novel deep network architecture for estimating the visibility directly from highway surveillance images. Specifically, we employ several image feature extraction methods to extract detailed structural, spectral, and scene depth features from the images. Next, we design a multi-scale fusion network to adaptively extract and fuse vital features for the purpose of estimating visibility. Furthermore, we create a real-scene dataset for model learning and performance evaluation. Our experiments demonstrate the superiority of our proposed method to the existing methods.

18F-FDG PET/CT-based deep learning radiomics predicts 5-years disease-free survival after failure to achieve pathologic complete response to neoadjuvant chemotherapy in breast cancer

BackgroundThis study aimed to assess whether a combined model incorporating radiomic and depth features extracted from PET/CT can predict disease-free survival (DFS) in patients who failed to achieve pathologic complete response (pCR) after neoadjuvant chemotherapy.ResultsThis study retrospectively included one hundred and five non-pCR patients. After a median follow-up of 71 months, 15 and 7 patients experienced recurrence and death, respectively. The primary tumor volume underwent feature extraction, yielding a total of 3644 radiomic features and 4096 depth features. The modeling procedure employed Cox regression for feature selection and utilized Cox proportional-hazards models to make predictions on DFS. Time-dependent receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) were utilized to evaluate and compare the predictive performance of different models. 2 clinical features (RCB, cT), 4 radiomic features, and 7 depth features were significant predictors of DFS and were included to develop models. The integrated model incorporating RCB, cT, and radiomic and depth features extracted from PET/CT images exhibited the highest accuracy for predicting 5-year DFS in the training (AUC 0.943) and the validation cohort (AUC 0.938).ConclusionThe integrated model combining radiomic and depth features extracted from PET/CT images can accurately predict 5-year DFS in non-pCR patients. It can help identify patients with a high risk of recurrence and strengthen adjuvant therapy to improve survival.

Soil dynamic response mapping and vulnerability analysis of earthquake hazard in Banda Aceh using HVSR method

Banda Aceh is a city with a high level of seismicity and earthquake risk. The seismicity level is mainly caused by two Sumatran Fault segments, i.e., the Aceh and Seulimeum Segments. Soft alluvium deposits aggravate the earthquake hazard in Banda Aceh. The presence of soft alluvium layers above the more rigid soil layer will cause earthquake wave amplification. This amplification phenomenon can increase the amplitude and energy of earthquake waves, making the earthquake more destructive.Therefore, dynamic response mapping and subsurface structure modeling are required to determine the soil characteristics of the research area. Such information is needed for the infrastructure and planning development of Banda Aceh City. This study obtained the subsurface model from microtremor data and a sequence of processing and modeling techniques. The process uses the Horizontal to Vertical Spectral Ratio method (HVSR). The resonance frequency and amplification values are identified from the ratio between the horizontal and vertical spectrum components depicted in the HVSR curves. Based on the obtained results, Banda Aceh has a predominant frequency values range of 0.54-11.72 Hz and an amplification range of 1.22-6.26. Low predominant frequency is mainly located in the study area’s southeast and center, which is associated with large sediment thickness. Meanwhile, high amplification values are mainly located on the study area’s southeast, middle, and northwest sides, which are associated with increased sediment depth and geological features in the form of swamp sediment. Then the value of seismic vulnerability (Kg ) can be obtained based on the predominant frequency and amplification value. In this study, the value of seismic vulnerability ranges between 0.72-31.77 area with relatively high vulnerability values in the middle, southeast, south, and southwest of the study area.

Indoor semantic segmentation based on Swin-Transformer

In recent years, with the rapid development of Transformer in the field of natural language processing, many researchers have realized its potential and gradually applied it to the field of computer vision, with a proliferation of theoretical approaches represented by Vision Transformer (ViT) and Data-efficient image Transformer (DeiT). On the basis of ViT, the famous Swin-Transformer was proposed as one of the best computer vision neural network backbones, which can be widely used in tasks such as image classification, target detection and video recognition. However, in the field of image segmentation, the semantic segmentation of indoor scenes is still very challenging due to the wide variety of objects, large differences in object sizes, and a large number of overlapping objects with occlusion. Aiming at the problem that the existing semantic segmentation of RGB-D indoor scenes cannot effectively fuse multimodal features, in this paper, we propose a novel indoor semantic segmentation algorithm based on Swin-Transformer. It attempts to apply Swin-Transformer to the field of indoor RGBD semantic segmentation, and tests the performance of the model by conducting extensive experiments on the mainstream indoor semantic segmentation datasets NYU-Depth V2 and SUN RGB-D. The experimental results show that the Swin-L RGB+Depth setting achieves 52.44% MIoU on the NYU-Depth V2 data and 51.15% MIoU on the SUN RGB-D data set, which reflects an excellent performance in the field of indoor semantic segmentation. The improved performance of the Depth features on the indoor semantic segmentation model has also been demonstrated in the experiments by controlling the type of input features. Our source code is publicly available at https://github.com/YunpingZheng/ISSSW.

Multi-scale split matching three-dimensional reconstruction method of surface microcracks on the silicon nitride bearing roller

Aiming at the characteristics of parametric discrete and non-significant depth of silicon nitride bearing roller surface microcracks. A multi-scale split matching method is proposed. To realize the full reconstruction of the characteristics of the three-dimensional morphology of the surface microcracks. Combined with the parametric discretization characteristics of the surface microcracks, a box filter is defined to do second-order Gaussian differentiation on the image. Construct a multiscale Hessian matrix in the scale space of the surface microcrack image. Determine the covariance-discrete feature points, and obtain the global covariance-discrete sparse point cloud of silicon nitride bearing roller surface microcracks by non-maximum value suppression. For the non-significant depth features of silicon nitride bearing roller surface microcracks. Dissect the surface microcrack distortion images. Divide each image into multiple regional pixel blocks and construct an aggregation function to determine the exact depth value of each pixel block at the cost of matching each pixel block with its surrounding pixel blocks. Dense point cloud reconstruction of non-significant depth features of silicon nitride bearing roller surface microcracks is realized. The average distribution of dense clusters of reprojection errors of the dense point cloud features is less than 0.4pix, the median residual of the field of view of the surface microcrack image is 0.57pix, and the total number of effectively extracted feature points is about 92.9 %. The detailed lossless recovery reconstruction of the parametric discrete, non-significant depth features of silicon nitride bearing roller surface microcracks is realized. A new idea is provided for the establishment of molecular dynamics geometric modeling of the surface microcrack extension process on the silicon nitride bearings roller.

Saliency-guided stairs detection on wearable RGB-D devices for visually impaired persons with Swin-Transformer

Accuracy stairs detection is crucial for people with visual impairment, as it can reduce the potential unforeseen risks of falling on stairs. Wearable RGB-D technology can assist blind and visually impaired individuals. However, existing stair detection algorithms on RGB-D images face difficulties in the stair material, texture, lighting, and direction. In this study, we proposed a saliency-guided stairs detection method based on Swin-Transformer to address the challenges mentioned above. First, saliency detection based on RGB-D images is used to learn spatial information for fast stair localization. Furthermore, we use the Swin-Transformer that incorporates key depth features of the stairs to solve orientation detection deficiencies. To evaluate the performance of our proposed method, we collected 3,290 RGB-D images, including the indoor and outdoor staircases. Experiments on our dataset show that our method can achieve high performance in terms of detection accuracy.

ISHS-Net: Single-View 3D Reconstruction by Fusing Features of Image and Shape Hierarchical Structures

The reconstruction of 3D shapes from a single view has been a longstanding challenge. Previous methods have primarily focused on learning either geometric features that depict overall shape contours but are insufficient for occluded regions, local features that capture details but cannot represent the complete structure, or structural features that encode part relationships but require predefined semantics. However, the fusion of geometric, local, and structural features has been lacking, leading to inaccurate reconstruction of shapes with occlusions or novel compositions. To address this issue, we propose a two-stage approach for achieving 3D shape reconstruction. In the first stage, we encode the hierarchical structure features of the 3D shape using an encoder-decoder network. In the second stage, we enhance the hierarchical structure features by fusing them with global and point features and feed the enhanced features into a signed distance function (SDF) prediction network to obtain rough SDF values. Using the camera pose, we project arbitrary 3D points in space onto different depth feature maps of the CNN and obtain their corresponding positions. Then, we concatenate the features of these corresponding positions together to form local features. These local features are also fed into the SDF prediction network to obtain fine-grained SDF values. By fusing the two sets of SDF values, we improve the accuracy of the model and enable it to reconstruct other object types with higher quality. Comparative experiments demonstrate that the proposed method outperforms state-of-the-art approaches in terms of accuracy.

Application of Multimedia Data Feature Extraction Technology in Teaching Classical Oil Painting

The cross-modal oil painting image generated by traditional methods makes it easy to miss the important information of the target part, and the generated image lacks realism. This paper combines the feature extraction technology of multimedia data with the generation confrontation network in deep learning, puts forward a generation model of classic oil painting, and applies it to university teaching. Firstly, the key frame extraction algorithm is used to extract the key frames in the video, and the channel attention network is introduced into the pre-trained ResNet-50 network to extract the static features of 2D images in short oil painting videos. Then, the depth feature mapping is carried out in the time dimension by using the double-stream I3D network, and the feature representation is enhanced by combining static and dynamic features. Finally, the high-dimensional features in the depth space are mapped to the two-dimensional space by using the opposition generation network to generate classic oil painting pictures.

CaP: Cardiovascular Disease Prediction using a Delta Layer based Center Vector Activation-centric DCNN

Cardiac disease stands as a primary contributor to mortality, representing a prevalent category of chronic and life-threatening conditions. Therefore, early detection is imperative. While existing research has sought to predict heart disease (HD) through Electrocardiogram (ECG) signals, there remains room for enhancement. This study introduces a novel approach for early HD detection based on the Delta Layer with Center Vector Activation-centric Deep Convolutional Neural Network (DLCVA-DCNN) within its research framework, namely: CaP. Initially, the input ECG signals undergo preprocessing using a Weighted Covariance Kalman Filter (WCKF) to eliminate noise. Subsequently, the preprocessed data is bifurcated: one branch transforms it into a binary image, while the other decomposes the signal to identify peak segments. The decomposition employs the Bivariate Ensemble Empirical Mode Decomposition (BEEMD) method, and the Pan-Tompkins Algorithm (PTA) is applied to ascertain the highest-frequency segments. The coupling information is then extracted from these peaks. Simultaneously, depth features are extracted from the binary image. The Linear Approximate Functional Walrus Optimization Algorithm (LAFWOA) is employed to select pertinent features from the coupling and depth features. These selected features are input into the DLCVA-DCNN classifier to discriminate disease and standard signals. The experimental analysis compares the proposed methodology with conventional frameworks based on performance metrics, revealing that the proposed approach achieves higher accuracy than existing techniques.

Multi-dimensional stereo face reconstruction for psychological assistant diagnosis in medical meta-universe

Virtual reality/augmented reality (VR/AR) technology is very important for meta-cosmic medicine. It is the technical link between the material world and the virtual world. The realization of face-to-face psychological diagnosis with three-dimensional holography is a difficult problem in meta-cosmic medicine. Because psychological diagnosis needs to observe rapidly changing microexpressions, it is difficult to capture them. At present, most of the research is based on two dimensions, which cannot meet the real-time reconstruction requirements of meta-universe medicine. To solve these problems, this paper introduces a 3D facial reconstruction algorithm composed of multiresolution feature fusion and attention mechanisms. It can be used for psychological assistant diagnosis of meta-cosmic medicine. First, face alignment and 3D coordinate construction of the feature point cloud are carried out through a 3D convolution coding network. After that, multiresolution cost convolution is used to extract and match the color, facial bones, and depth features of texture images. Finally, the transposed three-dimensional convolution is used to decode and output. To achieve the real-time reconstruction of psychotherapy, we designed a three-dimensional channel attention mechanism module for acceleration. This makes this method suitable for the construction of virtual doctors and virtual patients in the virtual platform of the medical universe of psychological medicine and can also complete the construction of the virtual teaching platform of the medical universe.

Classification of Alzheimer’s Disease Based on White Matter Connectivity Network

Alzheimer’s disease (AD) is one of the most common irreversible brain diseases in the elderly. Mild cognitive impairment (MCI) is an early symptom of AD, and the early intervention of MCI may slow down the progress of AD. However, due to the subtle neuroimaging differences between MCI and normal control (NC), the clinical diagnosis is subjective and easy to misdiagnose. Machine learning can extract depth features from neural images, and analyze and label them to assist the diagnosis of diseases. This paper combines diffusion tensor imaging (DTI) and support vector machine (SVM) to classify AD, MCI, and NC. First, the white matter connectivity network was constructed based on DTI. Second, the nodes with significant differences between groups were screened out by the two-sample t-test. Third, the optimal feature subset was selected as the classification feature by recursive feature elimination (RFE). Finally, the Gaussian kernel support vector machine was used for classification. The experiment tested and verified the data downloaded from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, and the area under the curve (AUC) of AD/MCI and MCI/NC are 0.94 and 0.95, respectively, which have certain competitive advantages compared with other methods.

LS-MVSNet: Lightweight self-supervised multi-view stereo

In recent years, significant progress has been made in multi-view stereo through the adoption of learning-based methods. However, current state-of-the-art approaches exhibit high computational and memory costs, owing to their reliance on semantic segmentation networks, neural rendering, and other advanced technologies for enhancing the accuracy of depth map prediction. To improve the efficiency of resource utilization and maintain accuracy in depth prediction, we present a self-supervised multi-view stereo framework that is lightweight and combines depth feature extraction modules, edge feature extraction modules, and data augmentation techniques. Specifically, the depth feature extraction module is employed to adaptively extract significant information from the input images for depth estimation. The edge feature extraction module is utilized to extract object contour features from the input images to augment the accuracy of depth inference at the edges. Experimental results on the DTU dataset show that our proposed lightweight self-supervised multi-view stereo framework, referred to as LS-MVSNet, outperforms the most advanced self-supervised MVS methods in terms of accuracy. Furthermore, extensive experiments on the Tanks and Temples dataset demonstrate the effective generalization ability of our model. The code will be released.

Bird's-Eye-View Semantic Segmentation With Two-Stream Compact Depth Transformation and Feature Rectification

Bird's-eye-view (BEV) perception has gained popularity since it provides a 3D world representation with scale consistency. Although existing camera-based solutions achieve excellent performance, the BEV positions related to features are still less accurate. In this paper, a BEV semantic segmentation framework with two-stream compact depth transformation and feature rectification is proposed. To balance the conflict that the feature maps ensemble tends to use two temporal frames with long interval, while shorter temporal frames are more beneficial to depth prediction, a two-stream compact depth transformation is designed. Between original temporal frames, we introduce an intermediate frame to decouple the joint depth estimation of original frames. The local representations of the intermediate frame are respectively matched with each original temporal frame to achieve stereo depth predictions, where compact cost volumes are built to significantly reduce memory usage with high discriminability in depth-dimension. Further, virtual camera intrinsic parameters are derived to realize adaption of compact cost volume to various 2D data augmentation and improve generalization. On this basis, BEV feature maps are obtained via feature transformation. With the influence of depth distribution errors to BEV feature map, a feature rectified segmentation network is proposed to dynamically adjust the position offsets of input features via deformable convolution and semantic information-guided feature learning. As a result, a dense and accurate BEV semantic map is obtained. In addition, a self-supervised depth estimation teacher is adopted to provide extra supervision for depth prediction of our segmentation framework. The effectiveness of the proposed method is verified on public datasets.