Keypoint Selection Research Articles

A deep learning model for FaceSwap and face-reenactment deepfakes detection

Recently, the higher availability of multimedia content on social websites, together with lightweight deep learning (DL) empowered tools like Generative Adversarial Networks (GANs) has caused the generation of realistic deepfakes. Such fabricated data has the potential to spread disinformation, revenge porn, initiate monetary scams, and can result in adverse immoral and illegal societal issues, etc. Hence, the accurate identification of deepfakes is mandatory to discriminate between real and manipulated content. In this work, we have presented a DL-based approach namely a unified network for FaceSwap (FS) and Face-Reenactment (FR) Deepfakes Detection (AUFF-Net). More clearly, both the spatial and temporal information from the video samples are used to detect two types of visual manipulations i.e., FS and FR. For this reason, a novel DL framework namely the Inception-Swish-ResNet-v2 model is introduced as a feature extractor for computing the information at the spatial level. While the Bi-LSTM model is utilized to measure the temporal information. Additionally, 3 dense layers are included at the last of the model structure to suggest a discriminative group of the feature vector We performed extensive experimentation on a challenging dataset namely the FaceForensic++, and attainede average accuracy values of 99.21 %, and 98.32 % for FS, or FR, respectively. Furthermore, we introduced an explainability module to show the reliable keypoints selection capability of our technique. Moreover, we have performed a cross-dataset evaluation to show the generalization power of our approach. Both the qualitative and quantitative results have confirmed the effectiveness of the suggested approach for visual manipulation categorization under the occurrence of various adversarial attacks.

Behaviour recognition of housed sheep based on spatio-temporal information

ABSTRACT Livestock behavior is related to the healthy breeding and welfare level. Therefore, monitoring the behavior of sheep is helpful to predict the health status of sheep and thus safeguard the production performance of sheep. Taking semi-housed sheep as the research object, a behavior identification method for housed sheep based on spatio-temporal information is proposed. Firstly, video acquisition of housed sheep is carried out, and sheep detection and tracking is implemented based on the YOLOv5 coupled with a Deep-SORT algorithm, which detects sheep from the flock and labels their identity information; then, sheep posture estimation is done based on the Alphapose algorithm, which estimates the keypoints in the skeleton; finally, the detected keypoints are inputted into the trained spatio-temporal graph convolutional network model, and the spatio-temporal graph constructed from the keypoints and edge information is used for sheep behavior recognition. In the study, two keypoints selection proposals were made to implement sheep behavior recognition for non-contacted detection of ruminating, lying down and other behaviors. The results show the feasibility of sheep behavior recognition based on spatio-temporal information in semi-housed farming. The method provides a new solution idea for intelligent monitoring of sheep behavior and health assessment.

BDR6D: Bidirectional Deep Residual Fusion Network for 6D Pose Estimation

Six-dimensional (6D) pose estimation is an important branch in the field of robotics focused on enhancing the ability of robots to manipulate and grasp objects. The latest research trend in 6D pose estimation is to directly predict the positions of two-dimensional (2D) keypoints from a single red, green, and blue (RGB) image through convolutional neural networks (CNNs) and establish a corresponding relationship with the three-dimensional (3D) keypoints of the model. Then, the perspective-n-point (PnP) algorithm is used to recover the 6D pose parameters. Currently, two challenges are encountered in pose estimation based on an RGB image. On the one hand, an RGB image lacks depth information, and it is thus difficult to directly obtain the corresponding geometric object information. On the other hand, when depth information is available, it is difficult to efficiently fuse the features of the RGB image with the features of the corresponding depth image. In this paper, we propose a bidirectional depth residual fusion network with a depth prediction (DP) network to estimate the 6D poses of objects (BDR6D). The BDR6D network predicts the depth information of objects using an RGB image, converts the depth information into point cloud information, and performs feature extraction and representation together with the RGB information during the feature extraction and representation stages. Specifically, the RGB image is fed into the BDR6D network, the DP network predicts the depth information of the objects in the image, and the depth map and RGB image are input into a point cloud network (PCN) and CNN, respectively, for feature extraction and representation. We build the bidirectional depth residual (BDR) structure so that the CNN and PCN can share information during feature extraction and representation. This approach allows the two networks to use each other’s local and global information to improve feature extraction and representation. For the keypoint selection stage, we propose an effective 2D keypoint selection method that considers the appearance and geometric information of the object of interest. We evaluate the proposed method with three benchmark datasets and compare it with other 6D pose estimation algorithms. The experimental results show that our method outperforms the state-of-the-art approach. Finally, we deploy our proposed method in conjunction with the Universal Robots 5 manipulator (UR5) robot to grasp and manipulate objects. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —The purpose of this paper is to solve the problem of 6D pose estimation for robot grasping. The existing RGB image-based pose estimation approach faces two challenges. On the one hand, a single RGB image lacks depth information, so that it is difficult to directly obtain the corresponding geometric object information. On the other hand, when depth information is available, it is difficult to efficiently fuse the features of the RGB image with the features of the corresponding depth image. To solve the above problems, a novel network that can predict the depth information of objects from an RGB image and fuse the depth information with the RGB information to estimate the 6D pose of objects is proposed. Furthermore, we propose an effective 2D keypoint selection method that considers the appearance and geometric information of objects of interest. We evaluate the proposed approach based on three benchmark datasets and the UR5 robot platform and verify that our method is effective.

Key-point based automated diagnosis for alveolar dehiscence in mandibular incisors using convolutional neural network

ObjectiveThe aim of this study was to propose an automated diagnosis method for alveolar dehiscence in the anterior teeth using Convolutional Neural Network (CNN). MethodsThe Cone-Beam Computed Tomography (CBCT) scanning was performed on 387 orthodontic patients at Shanghai Stomatological Hospital. A total of 1017 mandibular incisors with the largest labiolingual sectional images were obtained from the CBCT data. Among the 1017 incisor images, 371 specimens were diagnosed with alveolar dehiscence. We proposed two strategies of automated diagnosis methods for alveolar dehiscence. The first strategy (referred to as the Binary Classification Method (BCM)) was to take the task directly as a classic binary classification, and five classification networks (ResNet50, ResNet101, VGG16, AlexNet, MobileNet) were tested in this task. The second strategy (referred to as the Key-Point based Method (KPM)) was to use the CNN to search two key points (i.e., the Cement-Enamel Junction (CEJ) and the Alveolar Crest (AC)) firstly and then make a diagnosis according to the distance between the two key points. At the same time, we proposed an image preprocessing method for the approximate location of mandibular incisors and an improved key point selection method to avoid the problems of missed detection. In both CNN strategies, 90% of the mandibular incisor images were assigned to the training dataset, and the rest 10% were assigned to the testing dataset. ResultsThe BCM showed limited performance in the diagnosis of alveolar dehiscence, with diagnostic accuracy below 70% for all the five classification networks. The KPM showed superior diagnostic performance with an accuracy of 90.2%, a sensitivity of 86.2% and a specificity of 92.6%, respectively, in the testing dataset. The proposed image preprocessing procedures also played an essential role in the diagnosis process, significantly improving the diagnostic accuracy by 7.0% in KPM. ConclusionThe results proved that the diagnosis of alveolar dehiscence could be realized using convolutional neural network and the proposed KPM have the advantage of high accuracy and real-time performance. This study suggests that the key-point based convolutional neural network might have the potential for the diagnosis of alveolar dehiscence in the clinic.

Robot Pose Estimation and Normal Trajectory Generation on Curved Surface Using an Enhanced Non-Contact Approach.

The use of robots for machining operations has become very popular in the last few decades. However, the challenge of the robotic-based machining process, such as surface finishing on curved surfaces, still persists. Prior studies (non-contact- and contact-based) have their own limitations, such as fixture error and surface friction. To cope with these challenges, this study proposes an advanced technique for path correction and normal trajectory generation while tracking a curved workpiece's surface. Initially, a key-point selection approach is used to estimate a reference workpiece's coordinates using a depth measuring tool. This approach overcomes the fixture errors and enables the robot to track the desired path, i.e., where the surface normal trajectory is needed. Subsequently, this study employs an attached RGB-D camera on the end-effector of the robot for determining the depth and angle between the robot and the contact surface, which nullifies surface friction issues. The point cloud information of the contact surface is employed by the pose correction algorithm to guarantee the robot's perpendicularity and constant contact with the surface. The efficiency of the proposed technique is analyzed by carrying out several experimental trials using a 6 DOF robot manipulator. The results reveal a better normal trajectory generation than previous state-of-the-art research, with an average angle and depth error of 1.8 degrees and 0.4 mm.

Real-Time In-Sensor Slope Level-Crossing Sampling for Key Sampling Points Selection for Wearable and IoT Devices

This article presents a slope level-crossing sampling analog-to-digital converter (ADC) that selects key sampling points for quantization in real time during sensing. It only performs quantization for the turning points in the input analog waveform and provides quantization results of the selected sampling points and timestamps between the selected sampling points. When the input analog signal is sparse, the proposed method reduces digital output data throughput. The processing unit generates a dynamic prediction of the input signal as well as an upper threshold and a lower threshold to form a tracking window. A comparator compares the input signal with the upper and lower threshold to determine if the prediction is successful. Quantization is performed only on unsuccessful predicted sampling points, which are considered key sampling points. A counter records timestamps between the unsuccessful predictions, which are the selected key sampling points. The processing unit also includes a neighbor amplitude filter and a slope filter to further reduce the number of sampling points and data throughput when the input signal is associated with high-frequency, low-amplitude noise and high-amplitude, low-frequency baseline wandering. Reconstruction of the analog signal can be achieved using linear interpolation or polynomial interpolation. The system has been implemented and tested using off-the-shelf components. The simulation and experimental results show that the proposed system can reduce the data throughput and achieve a data compression ratio of 7.1 compared with a conventional successive approximation register (SAR) ADC with a 10-bit resolution when sampling an electrocardiogram (ECG) signal.

An Intelligent Auxiliary Framework for Bone Malignant Tumor Lesion Segmentation in Medical Image Analysis.

Bone malignant tumors are metastatic and aggressive, with poor treatment outcomes and prognosis. Rapid and accurate diagnosis is crucial for limb salvage and increasing the survival rate. There is a lack of research on deep learning to segment bone malignant tumor lesions in medical images with complex backgrounds and blurred boundaries. Therefore, we propose a new intelligent auxiliary framework for the medical image segmentation of bone malignant tumor lesions, which consists of a supervised edge-attention guidance segmentation network (SEAGNET). We design a boundary key points selection module to supervise the learning of edge attention in the model to retain fine-grained edge feature information. We precisely locate malignant tumors by instance segmentation networks while extracting feature maps of tumor lesions in medical images. The rich contextual-dependent information in the feature map is captured by mixed attention to better understand the uncertainty and ambiguity of the boundary, and edge attention learning is used to guide the segmentation network to focus on the fuzzy boundary of the tumor region. We implement extensive experiments on real-world medical data to validate our model. It validates the superiority of our method over the latest segmentation methods, achieving the best performance in terms of the Dice similarity coefficient (0.967), precision (0.968), and accuracy (0.996). The results prove the important contribution of the framework in assisting doctors to improve the accuracy of diagnosis and clinical efficiency.

Новые алгоритмы применения логических операций для выделения объектов железнодорожной инфраструктуры

The article presents new algorithms for improving the segmentation of railway infrastructure images using logical operations. The main focus is on the development of efficient methods for highlighting objects, such as trains, wagons, platforms, and railway signals in images. The proposed methods helpmake it possible to efficiently process images by removing noise, highlighting objects, and improving the quality of recognition. The results of the experiments confirm that the new algorithms can achieve an improvement in accuracy by an average value of 18 % compared to the method of active contours method with wavelet by the Haar wavelet transformation and by 14 % compared to algorithms using keypoint selection.

ACKSNet: adaptive center keypoint selection for object detection

ACKSNet: adaptive center keypoint selection for object detection

DeepFocus: A visual focus of attention detection framework using deep learning in multi-object scenarios

In recent years, recognizing the visual focus of attention (VFoA) has attracted much attention among computer vision experts due to its various Human–Computer Interaction (HCI) or Human-Robot Interaction (HRI) applications. Although eye gaze is a potential cue to determine someone’s focus of attention (FOA), it is challenging to determine FOA alone when the interacting partners are far away or the camera cannot capture high-resolution images from long-distance. Therefore, the head pose can be used as an approximation to recognize the focus of someone’s attention. This paper proposes a vision-based framework to detect the FOA of humans using nine head poses consisting of four main modules: face detection and facial key-point selection (FDKPSM), head pose classification (HPCM), object localization and classification (OLCM), and focus of attention estimation (FoAEM). The FDKPSM uses the Multi-task Cascaded Neural Network (MTCNN) framework to detect head poses, and the HPCM classifies them into nine classes using the ResNet18. To estimate the FoA, the FoAEM uses a mapping Algorithm (EFoA) which integrates head poses on the focused object. Experimental results show that the proposed model outperformed other deep learning models by achieving the highest accuracy on three datasets: BIWI-M (96.97%), Pointing’04-M (96.04%) and HPoD 9 (98.99%). The visual focus of the attention model gained an accuracy of 94.12% in the multi-object scenario.

Learning-Based Pose Estimation of Non-Cooperative Spacecrafts with Uncertainty Prediction

Estimation of spacecraft pose is essential for many space missions, such as formation flying, rendezvous, docking, repair, and space debris removal. We propose a learning-based method with uncertainty prediction to estimate the pose of a spacecraft from a monocular image. We first used a spacecraft detection network (SDN) to crop out the rectangular area in the original image where only spacecraft exist. A keypoint detection network (KDN) was then used to detect 11 pre-selected keypoints with obvious features from the cropped image and predict uncertainty. We propose a keypoints selection strategy to automatically select keypoints with higher detection accuracy from all detected keypoints. These selective keypoints were used to estimate the 6D pose of the spacecraft with the EPnP algorithm. We evaluated our method on the SPEED dataset. The experiments showed that our method outperforms heatmap-based and regression-based methods, and our effective uncertainty prediction can increase the final precision of the pose estimation.

Geo-Location Method for Images of Damaged Roads

Due to the large difference between normal conditions and damaged road images, geo-location in damaged areas often fails due to occlusion or damage to buildings and iconic signage in the image. In order to study the influence of post-war building and landmark damage conditions on the geolocation results of localization algorithms, and to improve the geolocation effect of such algorithms under damaged conditions, this paper used informative reference images and key point selection. Aiming at the negative effects of occlusion and landmark building damage in the retrieval process, a retrieval method called reliability- and repeatability-based deep learning feature points is proposed. In order to verify the effectiveness of the above algorithm, this paper constructed a data set consisting of urban, rural and technological parks and other road segments as a training set to generate a database. It consists of 11,896 reference images. Considering the cost of damaged landmarks, an artificially generated method is used to construct images of damaged landmarks with different damage ratios as a test set. Experiments show that the database optimization method can effectively compress the storage capacity of the feature index and can also speed up the positioning speed without affecting the accuracy rate. The proposed image retrieval method optimizes feature points and feature indices to make them reliable against damaged terrain and images. The improved algorithm improved the accuracy of geo-location for damaged roads, and the method based on deep learning has a higher effect on the geo-location of damaged roads than the traditional algorithm. Furthermore, we fully demonstrated the effectiveness of our proposed method by constructing a multi-segment road image dataset.

Object aspect classification and 6DoF pose estimation

• We introduce a simple aspect-representation for 3D objects for 6DoF pose estimation. • This representation is used to cluster appearances of an object as aspects. • Proposed model predicts both the aspect of a given object and the semantic-key points. • Multi-tasking to high confidence key-point vectors hence more accurate results. • Proposed method is significantly faster due to small backbone and limited PnP Space. We present a novel approach to the problem of estimating a given object's 6DoF pose from a single RGB image. Recent works focus on a multi-stage approach, which first detects key-points followed by perspective -n-point pose estimation algorithm and a pose refinement procedure. We show that adding a classifier block estimating the predefined aspects of the objects improves the multi-stage process. This is due to the fact that the additional classifier acts as a constraint simplifying the required neural network and at the same time yielding better key-point selection. We reduce the search space for the key-point selection and exclude false-positives by mapping the appearance of an object to an aspect. The simplified neural network allows faster inference and a smaller footprint. Our experiments show that our hypothesis performs similar to the state-of-the-art on three different datasets. We also show that an off-the-shelf refinement process can further improve our results to surpass state-of-the-art on several objects. Another advantage is, the proposed pipeline can run efficiently on real-time due to the smaller neural network backbone used. The code to replicate this research will be publicly available at https://github.com/greymad/6DoFPoseAspects

Structure boundary-preserving U-Net for prostate ultrasound image segmentation.

Prostate cancer diagnosis is performed under ultrasound-guided puncture for pathological cell extraction. However, determining accurate prostate location remains a challenge from two aspects: (1) prostate boundary in ultrasound images is always ambiguous; (2) the delineation of radiologists always occupies multiple pixels, leading to many disturbing points around the actual contour. We proposed a boundary structure-preserving U-Net (BSP U-Net) in this paper to achieve precise prostate contour. BSP U-Net incorporates prostate shape prior to traditional U-Net. The prior shape is built by the key point selection module, which is an active shape model-based method. Then, the module plugs into the traditional U-Net structure network to achieve prostate segmentation. The experiments were conducted on two datasets: PH2 + ISBI 2016 challenge and our private prostate ultrasound dataset. The results on PH2 + ISBI 2016 challenge achieved a Dice similarity coefficient (DSC) of 95.94% and a Jaccard coefficient (JC) of 88.58%. The results of prostate contour based on our method achieved a higher pixel accuracy of 97.05%, a mean intersection over union of 93.65%, a DSC of 92.54%, and a JC of 93.16%. The experimental results show that the proposed BSP U-Net has good performance on PH2 + ISBI 2016 challenge and prostate ultrasound image segmentation and outperforms other state-of-the-art methods.

Robust data hiding scheme through distinct keypoint selection exploiting modified Bilateral-Laplacian SIFT with encoding pipeline

SIFT (Scale Invariant Feature Transform) has recently been used to secure multimedia materials from geometric and non-geometric attacks. SIFT is commonly used in Computer Vision to identify and describe local features in images using the Difference of Gaussian (DoG) method, which has a high processing cost and a strong affinity towards the edge, resulting in a low contrast point. By introducing modified SIFT (mSIFT), which selects the most distinguishable key point with their feature descriptors, an attempt has been made to construct a rotation and scaling resistant data hiding strategy in this study. Unlike SIFT, a Bilateral-Laplacian filter was employed to determine the extrema, which took into account both geometric and photometric distance to reduce noise and retain the edge. This adds to the benefits of the proposed scheme’s resistance against geometric attacks, even when the tampering rate is large. Furthermore, during data embedding, the suggested technique achieves excellent security by utilizing Arnold chaotic function, pseudorandom sequence generator, and a newly created unique encoding pipeline. The proposed technique is secure and resistant to various steganographic attacks while preserving good visual quality, according to experimental results and analysis, as well as comparisons with existing state-of-the-art methods.

Learning Dense Features for Point Cloud Registration Using a Graph Attention Network

Point cloud registration is a fundamental task in many applications such as localization, mapping, tracking, and reconstruction. Successful registration relies on extracting robust and discriminative geometric features. Though existing learning-based methods require high computing capacity for processing a large number of raw points at the same time, computational capacity limitation is not an issue thanks to powerful parallel computing process using GPU. In this paper, we introduce a framework that efficiently and economically extracts dense features using a graph attention network for point cloud matching and registration (DFGAT). The detector of the DFGAT is responsible for finding highly reliable key points in large raw data sets. The descriptor of the DFGAT takes these keypoints combined with their neighbors to extract invariant density features in preparation for the matching. The graph attention network (GAT) uses the attention mechanism that enriches the relationships between point clouds. Finally, we consider this as an optimal transport problem and use the Sinkhorn algorithm to find positive and negative matches. We perform thorough tests on the KITTI dataset and evaluate the effectiveness of this approach. The results show that this method with the efficiently compact keypoint selection and description can achieve the best performance matching metrics and reach the highest success ratio of 99.88% registration in comparison with other state-of-the-art approaches.

A Two-Stage Correspondence-Free Algorithm for Partially Overlapping Point Cloud Registration.

Point cloud registration is a key task in the fields of 3D reconstruction and automatic driving. In recent years, many learning-based registration methods have been proposed and have higher precision and robustness compared to traditional methods. Correspondence-based learning methods often require that the source point cloud and the target point cloud have homogeneous density, the aim of which is to extract reliable key points. However, the sparsity, low overlap rate and random distribution of real data make it more difficult to establish accurate and stable correspondences. Global feature-based methods do not rely on the selection of key points and are highly robust to noise. However, these methods are often easily disturbed by non-overlapping regions. To solve this problem, we propose a two-stage partially overlapping point cloud registration method. Specifically, we first utilize the structural information and feature information interaction of point clouds to predict the overlapping regions, which can weaken the impact of non-overlapping regions in global features. Then, we combine PointNet and the self-attention mechanism and connect features at different levels to efficiently capture global information. The experimental results show that the proposed method has higher accuracy and robustness than similar existing methods.

Computed Tomography Image Analysis of Body Fat Based on Multi-Image Information.

Body fat assessment is required as part of an objective health assessment, both for nonobese and obese people. Image-based body fat assessment will enable faster diagnosis. Body fat analysis that accounts for age and sex will help in both diagnosis and correlating diseases and fat distribution. After evaluating computed tomography imaging algorithms to identify and segment human abdominal and subcutaneous fat, we present an improved region growing scale-invariant feature transform algorithm. It applies Naive Bayes image thresholding for key point selection and image matching. This method enables rapid and accurate comparison and matching of images from multiple databases and improves the efficiency of image processing.

Factors influencing anesthesiologistsperceived usefulness of ACRM (Anaesthesia Crisis Resources Management) key points after a combined ACLS (AHA)-ACRM similation training

Background: The factors that may contribute to learners’ perception about the usefulness of Anaesthesia Crisis Resource Management (ACRM) key points are little known. Objectives: We investigated the link between demographic factors and the effect of an ACRM simulation training on anaesthesiologists’ perceived value of ACRM key points. Design: A prospective pre- and post-test survey from December 2017 to December 2019 of 111 anaesthesiologists involved into a combined ACRM-ACLS course in a simulation centre. Methods: Before and after the course participants were asked to indicate which were, in their opinion, the 5 ACRM key points most relevant for managing an anaesthetic emergency. No taxonomy tool of the 15 ACRM key points was used for teaching purposes. Main outcome measures: Pre-/post-course differences in participants’ subjective choices were connected by logistic regression analysis with demographic factors which included age, gender, years of work as anaesthesiologist, the amount of updating exposure, and familiarity with ACRM. Results: In median participants (47 M/64 F) had an age of 42 years (IQR 34-55 years), and 10 years working experience as anaesthesiologists (IQR 4-20 years). Around 20% of them had never heard of ACRM prior to this course. Communication was selected by up to 75% of participants without pre-/post-course differences. Although the other 4 ACRM points remained heterogeneously selected even after the course, we observed post-course vs pre-course increase in the selection rate of the ACRM points that address leadership, correct distribution of workload and utilization of all available resources. Among participants’ characteristics, the lack of familiarity with ACRM was the only significant predictor of the number of pre- to post-course changes in ACRM key points selection (OR=3.03, CI 95% 1.04 -9.09; p=0.0418). Conclusions: The familiarity with ACRM should be considered when planning ACRM training, especially in cases where the ACRM training is not yet part of a formal education in anaesthesia.

Stable keypoints selection for 2D LiDAR based place recognition with map data reduction

AbstractThis paper presents a new feature based approach for place recognition using 2D LiDAR (Light Detection And Ranging) data. The main contribution lies in the mapping process. It includes a keypoint selection strategy to model places with persistent keypoints from concatenated LiDAR scans. Our objective is to achieve map data reduction while maintaining good place recognition performace. LiDAR scans are concatenated with a registration algorithm and keypoints are extracted from each scan. Based on a regular grid, our approach measures the occurrence of similar keypoints in each region of interest defined by a grid cell. Only keypoints with occurrences beyond a threshold, qualified as stable keypoints, are kept in the place model called submap. The environment is therefore modeled as a collection of submaps, which constitutes the global map. Place recognition consists in submap matching followed by a two steps geometric verification. In the first stage, optimal parameters are set using corrected data. Mapping parameters satisfy six criteria among which is the spatial distribution distance, which represents another contribution of our work. It gives a measure on how well keypoints are distributed in space. Place recognition optimal parameters are set through global localization. In the second stage, the approach is evaluated using raw data in the contexts of global localization, and loop closure detection in a SLAM framework. The results obtained using popular real data sets show that our approach achieves significant map data reduction (up to $92\%$ ) while maintaining good place recognition performance, comparable to state-of-the-art methods.