Perceiving threats is an innate human instinct. During driving, humans naturally focus their attention on objects that pose real potential risks. Motivated by this observation, we shift the focus from traditional class-based detection to a novel task termed threat-oriented reasoning detection in autonomous driving. This task aims to localize threat objects and reason about their threat levels from a driver-centric perspective. To support this task, we build a benchmark comprising diverse corner-case scenarios, annotated by multiple experienced drivers to reflect human-aligned threat cognition. Given the reasoning demands of this task, we then explore the capabilities of multi-modal large language models (MLLMs) and introduce two methods based on whether the MLLM supports object detection: 1) For MLLMs lacking detection capability, we introduce ThreatCoT, a plug-and-play training-free method that combines chain-of-thought (CoT) with a visual expert toolchain to support step-by-step reasoning. 2) For MLLMs with detection support, we introduce ThreatReasoner, an end-to-end reinforcement learning (RL)-based method built on the GRPO algorithm, which enables per-object reasoning through a fully unsupervised reward strategy. Both quantitative and qualitative experiments show that our methods can effectively unlock the new capabilities of MLLM in threat-oriented reasoning detection.

Lane line detection in complex scenarios is a key part of the intelligent driving environment sensing technology, essential for ensuring traffic safety and improving transportation system efficiency. However, existing models ignore the data imbalance of the far-end, near-end, foreground object and background image, it leads to the inconsistency between the detected curve lane line and reality, and the poor complementarity of different scales of information. To solve the above problems, a high-precision and intelligent lane line detection algorithm based on bow height feature points (BHFP) and cross-scale feature correlation network (CFCNet) is proposed, the sub-module functions are shown below. A key point regression method based on BHFP extraction and box intersection over the union loss function correction was used to enhance the automatic modeling ability of the curve lane line. The CFCNet is used to mine the similarity and different characteristics of backbone network output data at different scales. A global information reflow enhancement model is used to highlight the advantages of deep and shallow features in correcting elongated target locations and sharpening structural details. The experimental results show that this algorithm improves the robustness and positioning efficiency of the lane line detection in complex road scenarios, and reduces the missing detection phenomenon of the model. It has significant advantages in the regression localization of lane line far-end targets.

Occupancy prediction plays a pivotal role in autonomous driving (AD) due to its capabilities of fine-grained 3D perception and general object recognition. However, existing methods often incur high computational costs, which conflict with AD's real-time demand. To this end, we redirect the focus from accuracy only to both accuracy and efficiency. By conducting a head-to-head comparison of existing methods, we find it challenging to balance accuracy and efficiency. We identify a core issue for this challenge: the strong coupling between geometry and semantics. Specifically, the predicted geometric structure (e.g., depth) guides the projection of 2D image features into 3D voxel space, which significantly affects feature discriminability and subsequent semantic learning. To address this issue, we focus on two key aspects: model design and learning strategies. 1) For model design, we propose a dual-branch network that disentangles the representation of geometry and semantics. The voxel branch utilizes a novel re-parameterized large-kernel 3D convolution to refine geometric structure efficiently, while the BEV branch employs temporal fusion and BEV encoding for efficient semantic learning. 2) For learning strategies, we propose to separate geometric learning from semantic learning by the mixup of ground-truth and predicted depths. Our method achieves 39.4% mIoU at 20 FPS on Occ3D-nuScenes, showcasing a state-of-the-art balance between accuracy and efficiency.

The image compression model has long struggled with adaptability and generalization, as the decoded bitstream typically serves only human or machine needs and fails to preserve information for unseen visual tasks. Therefore, this paper innovatively introduces supervision obtained from multimodal pre-training models and incorporates adaptive multi-objective optimization tailored to support both human visual perception and machine vision simultaneously with a single bitstream, denoted as Unified and Generalized Image Coding for Machine (UG-ICM). Specifically, to get rid of the reliance between compression models with downstream task supervision, we introduce Contrastive Language-Image Pre-training (CLIP) models into the training constraint for improved generalization. Global-to-instance-wise CLIP supervision is applied to help obtain hierarchical semantics that make models more generalizable for the tasks relying on the information of different granularity. Furthermore, for supporting both human and machine visions with only a unifying bitstream, we incorporate a conditional decoding strategy that takes as conditions human or machine preferences, enabling the bitstream to be decoded into different versions for corresponding preferences. As such, our proposed UG-ICM is fully trained in a self-supervised manner, i.e., without awareness of any specific downstream models and tasks. The extensive experiments have shown that the proposed UG-ICM is capable of achieving remarkable improvements in various unseen machine analytics tasks, while simultaneously providing perceptually satisfying images.

Remote Sensing Image (RSI) segmentation has made significant strides, emerging as a leading solution for interpreting remote sensing data. However, due to the substantial domain gap between different remote sensors and limited computational resources, existing RSI segmentation methods often suffer from poor generalization performance. To address these challenges, we propose a COst-effective and Whole-process Domain Adaptation solution, namely COWDA, which adapts models at both the train and test time through three key phases: 1) Source-data Domain Alignment: We employ traditional image stylization techniques to translate source images into target-style alternatives, which avoids the computationally intensive need for auxiliary neural network models. 2) Target-data Train-time Fine-tuning: We propose a joint positive and negative learning (JPNL) algorithm that adds both positive and negative samples to effectively learn domain-invariant knowledge from noisy pseudo-labeled target data. 3) Test-time Adaptation: We propose an entropy-weighted test-time adaptation strategy to update the trained model with online test samples, further enhancing its performance. Extensive experiments on two widely-used domain adaptation benchmarks for remote sensing show that COWDA improves state-of-the-art counterparts by 1.4% and 2.6% F1 scores, respectively.

Self-Training has proven to be a simple yet effective framework in source-free domain adaptation (SFDA) for medical image segmentation. However, prevalent domain discrepancies and privacy restrictions on source domain data can lead to the negative optimization problem. Current methods predominantly focus on rigorous pseudo-label quality assessment, yet they often overlook the specific prior knowledge inherent to medical images and the potential of online self-training to combat negative optimization. To address these gaps, this paper introduces a selective online self-training method, tackling the issue from both the pseudo-label selection and learning stages. In the selection stage, we identify a domain-invariant prior related to organ shape, and propose a class-prior guided selection mechanism to mitigate class imbalance in pseudo-labels. In the learning stage, we introduce a strategy to selectively update the teacher model based on evolutionary state feedback generated by the student model. Extensive experiments conducted on two widely used benchmarks show the effectiveness of our method, which achieves notable 2.0% and 4.1% Dice improvements.

Although RAW images offer advantages over sRGB by avoiding ISP-induced distortion and preserving more information in low-light conditions, their widespread use is limited due to high storage costs, transmission burdens, and the need for significant architectural changes for downstream tasks. To address the issues, this paper explores a new raw-based machine vision paradigm, termed Compact RAW Metadata-guided Image Refinement (CRM-IR). In particular, we propose a Machine Vision-oriented Image Refinement (MV-IR) module that refines sRGB images to better suit machine vision preferences, guided by learned raw metadata. Such a design allows the CRM-IR to focus on extracting the most essential metadata from raw images to support downstream machine vision tasks, while remaining plug-and-play and fully compatible with existing imaging pipelines, without any changes to model architectures or ISP modules. We implement our CRM-IR scheme on various object detection networks, and extensive experiments under low-light conditions demonstrate that it can significantly improve performance with an additional bitrate cost of less than $10^{-3}$ bits per pixel.

The two-scale model (TSM) has been widely used to calculate scattering from surfaces with multiple roughness scales. In our study, we derived an improved TSM based on the angular composite model (ACM), which expresses the calculation of the scattering coefficient as an integral over the surface angles, making it possible to accelerate the process by combining with Monte Carlo integration. In addition, any method can be employed to calculate the small-scale scattering. We use the first-order small-slope approximation model (SSA1) with wider applicability instead of the small perturbation model (SPM) to calculate the scattering contribution from small-scale rough surfaces. We analyze the roughness of short waves on anisotropic wind-driven sea surfaces, give numerical examples of horizontally polarized backscattering and bistatic scattering, taking into account the L-band and different wind speeds, and compare them with the measurement data, classic TSM (KA + SPM), facet-based TSM (KA + SSA1), closed-form two-scale model [bistatic anisotropic polarimetric TSM (BA-PTSM)], and second-order small-slope approximation model (SSA2) algorithms. The proposed bistatic scattering model is in general agreement with the classic model and is demonstrated to be more consistent with measurements at larger scattering angles and more insensitive to the cutoff wavenumber.

Given the widespread dissemination of digital audio and the advancements in speech synthesis technologies, protecting audio copyright has become a critical issue. Although watermarks play an important role in copyright verification and forensic analysis, they are insufficient to proactively defend against malicious speech synthesis. To address this issue, we introduce a novel adversarial speech synthesis watermarking mechanism (ASSMark), which simultaneously traces the audio copyright and disrupts the speech synthesis models by embedding robust adversarial watermarks in a one-time manner. Specifically, we design a unified training framework that models the embedding of watermarks and adversarial perturbations as collaborative tasks. This approach allows for the fine-tuning of any robust watermark into an adversarial watermark, resulting in watermarked audio that can effectively defend against unauthorized speech synthesis attacks. Experimental results demonstrate that ASSMark achieves over 90% protection rate even to unknown black-box models. Compared to simplistic two-step protection methods, it not only effectively resists synthesis attacks but also achieves superior watermark extraction accuracy and speech quality, offering an outstanding solution for protecting audio copyright.

Open World Object Detection (OWOD) aims to adapt object detection to an open-world environment, so as to detect unknown objects and learn knowledge incrementally. Existing OWOD methods typically leverage training sets with a relatively small number of known objects. Due to the absence of generic object knowledge, they fail to comprehensively perceive objects beyond the scope of training sets. Recent advancements in large vision models (LVMs), trained on extensive large-scale data, offer a promising opportunity to harness rich generic knowledge for the fundamental advancement of OWOD. Motivated by Segment Anything Model (SAM), a prominent LVM lauded for its exceptional ability to segment generic objects, we first demonstrate the possibility to employ SAM for OWOD and establish the very first SAM-Guided OWOD baseline solution. Subsequently, we identify and address two fundamental challenges in SAM-Guided OWOD and propose a pioneering SAM-Guided Robust Open-world Detector (SGROD) method, which can significantly improve the recall of unknown objects without losing the precision on known objects. Specifically, the two challenges in SAM-Guided OWOD include: 1) Noisy labels caused by the class-agnostic nature of SAM; 2) Precision degradation on known objects when more unknown objects are recalled. For the first problem, we propose a dynamic label assignment (DLA) method that adaptively selects confident labels from SAM during training, evidently reducing the noise impact. For the second problem, we introduce cross-layer learning (CLL) and SAM-based negative sampling (SNS), which enable SGROD to avoid precision loss by learning robust decision boundaries of objectness and classification. Experiments on public datasets show that SGROD not only improves the recall of unknown objects by a large margin (~20%), but also preserves highly-competitive precision on known objects. The program codes are available at https://github.com/harrylin-hyl/SGROD.

High Dynamic Range (HDR) images present unique challenges for Learned Image Compression (LIC) due to their complex domain distribution compared to Low Dynamic Range (LDR) images. In coding practice, HDR-oriented LIC typically adopts preprocessing steps (e.g., perceptual quantization and tone mapping operation) to align the distributions between LDR and HDR images, which inevitably comes at the expense of perceptual quality. To address this challenge, we rethink the HDR imaging process which involves fusing multiple exposure LDR images to create an HDR image and propose a novel HDR image compression paradigm, Unifying Imaging and Compression (HDR-UIC). The key innovation lies in establishing a seamless pipeline from image capture to delivery and enabling end-to-end training and optimization. Specifically, a Mixture-ATtention (MAT)-based compression backbone merges LDR features while simultaneously generating a compact representation. Meanwhile, the Reference-guided Misalignment-aware feature Enhancement (RME) module mitigates ghosting artifacts caused by misalignment in the LDR branches, maintaining fidelity without introducing additional information. Furthermore, we introduce an Appearance Redundancy Removal (ARR) module to optimize coding resource allocation among LDR features, thereby enhancing the final HDR compression performance. Extensive experimental results demonstrate the efficacy of our approach, showing significant improvements over existing state-of-the-art HDR compression schemes. Our code is available at: https://github.com/plf1999/HDR-UIC.

Spatiotemporal traffic data imputation is a fundamental task in numerous smart mobility applications. Existing studies indicate that accurately estimating missing values from observed data relies on capturing the spatiotemporal dependencies in traffic data. However, such dependencies may exhibit distinct characteristics across varying spatiotemporal areas, such as local short-term traffic fluctuations versus global long-range periodic commuting patterns. The comprehensive multiscale nature of dependencies in traffic data, encompassing more than just local and global levels, has not been well explored in the literature. To address this issue, we propose a physics-regularized multiscale attention network (PRMAN) that hierarchically extracts spatiotemporal features from local dynamics to global trends. Specifically, the proposed PRMAN builds upon the novel Swin Transformer and introduces a hierarchical architecture that performs self-attention in local spatiotemporal windows. By systematically expanding the window size across layers, this hierarchical design explicitly addresses the distinct characteristics between local and global spatiotemporal dependencies at different scales. Meanwhile, a physics-regularized loss function is developed to align learned spatiotemporal dependencies with traffic dynamics described by the fundamental diagram. This improves the model’s generalizability beyond the training data, ensuring robust performance on unseen datasets. Numerical experiments on multiple benchmark datasets demonstrate that our proposed PRMAN achieves state-of-the-art performance in handling diverse and complex missing data patterns. The code and model are publicly available athttps://github.com/2222ad/PRMAN

Image Coding for Machine (ICM) aims to compress an image so that the reconstructed one can meet the requirements of both human vision and machine vision. Existing methods apply the constraint from the downstream models to improve machine analytics performance while compromising the visual quality. This paper proposes a novel adversarially augmented adaptation route that achieves a better trade-off between the utility of the human and machine perspectives by making slight changes to the image manifold. In detail, a targeted adversarial attack is employed to generate subtle image perturbations that are nearly imperceptible to humans but significantly improve machine analytic performance. These perturbed images would be subsequently employed as ground truth to guide training/fine-tuning of an end-to-end image compression network. Note that, our method is a plug-and-play framework that does not rely on any change in existing architecture or loss functions. Extensive experimental results demonstrate the superiority of the proposed scheme over conventional ICM frameworks and the effectiveness of our design.

Pseudo-labeling has proven to be a simple yet effective technique for semi-supervised object detection (SSOD). However, the inevitable noise problem in pseudo-labels seriously hinders SSOD methods. Existing methods primarily focus on classification noise, while the specific and non-negligible localization noise remains not well-addressed. This paper analyzes the localization noise arising from the alternating learning and generation phases. For the generation phase, we innovatively explore the self-correction ability of models, stepping beyond the simple pseudo-label selection. We propose a localization pseudo-label correction (LPC) strategy to self-correct pseudo boxes and enhance prediction stability. In the learning phase, we propose a noisy localization loss (NLL) to enlarge the penalty of inconsistent predictions, thereby improving localization accuracy. Applied to two classic SSOD methods (Soft Teacher and Unbiased Teacher) and a recent state-of-the-art method (PseCo), our approach consistently improves accuracy across all of them.

Full-reference image quality assessment (FR-IQA) models generally operate by measuring the visual differences between a degraded image and its reference. However, existing FR-IQA models including both the classical ones (e.g., PSNR and SSIM) and deep-learning based measures (e.g., LPIPS and DISTS) still exhibit limitations in capturing the full perception characteristics of the human visual system (HVS). In this paper, instead of designing a new FR-IQA measure, we aim to explore a generalized human visual attention estimation strategy to mimic the process of human quality rating and enhance existing IQA models. In particular, we model human attention generation by measuring the statistical dependency between the degraded image and the reference image. The dependency is captured in a training-free manner by our proposed sliced maximal information coefficient and exhibits surprising generalization in different IQA measures. Experimental results verify the performance of existing IQA models can be consistently improved when our attention module is incorporated. The source code is available at https://github.com/KANGX99/SMIC.

Few-Shot learning deals with learning a model capable of recognizing new classes when provided with limited labeled data. Recently, Contrastive Language-Image Pre-training (CLIP) based methods have shown significant potential in this filed. However, due to the equal treatment for textual prompts, existing CLIP-based methods fail in establishing a strong text classifier, consequently limiting their performance. These textual prompts are obtained by querying large language models (LLMs), making it unreasonable to assume uniform quality across all of them. To address this issue, this paper proposes an Image-guided Text weighting (IGTW) module, which employs the feature similarity of training images and textual prompts to guide the weighting of textual prompts. After applying our method to the recent state-of-the-art method (i.e., CaFo) and classic method (i.e., Tip-Adapter), consistent improvements are achieved across 11 few-shot learning datasets, proving the effectiveness and universality of our method.

Autonomous driving (AD) is a typical application that requires effectively exploiting multimedia information. For AD, it is critical to ensure safety by detecting unknown objects in an open world, driving the demand for open world object detection (OWOD). However, existing OWOD methods treat generic objects beyond known classes in the train set as unknown objects and prioritize recall in evaluation. This encourages excessive false positives and endangers safety of AD. To address this issue, we restrict the definition of unknown objects to threatening objects in AD, and introduce a new evaluation protocol, which is built upon a new metric named U-ARecall, to alleviate biased evaluation caused by neglecting false positives. Under the new evaluation protocol, we re-evaluate existing OWOD methods and discover that they typically perform poorly in AD. Then, we propose a novel OWOD paradigm for AD based on fine-tuning foundational open-vocabulary models (OVMs), as they can exploit rich linguistic and visual prior knowledge for OWOD. Following this new paradigm, we propose a brand-new OWOD solution, which effectively addresses two core challenges of fine-tuning OVMs via two novel techniques: 1) the maintenance of open-world generic knowledge by a dual-branch architecture; 2) the acquisition of scenario-specific knowledge by the visual-oriented contrastive learning scheme. Besides, a dual-branch prediction fusion module is proposed to avoid post-processing and hand-crafted heuristics. Extensive experiments show that our proposed method not only surpasses classic OWOD methods in unknown object detection by a large margin (∼× U-ARecall), but also notably outperforms OVMs without fine-tuning in known object detection (∼ 20% K-mAP). Our codes are available at https://github.com/harrylin-hyl/AD-OWOD.

Summary Clustering ensemble is a popular approach for identifying data clusters that combines the clustering results from multiple base clustering algorithms to produce more accurate and robust data clusters. However, the performance of clustering ensemble algorithms is highly dependent on the quality of clustering members. To address this problem, this paper proposes a member enhancement‐based clustering ensemble (MECE) algorithm that selects the ensemble members by considering their distribution consistency. MECE has two main components, called heterocluster splitting and homocluster merging. The first component estimates two probability density functions (p.d.f.s) estimated on the sample points of an heterocluster and represents them using a Gaussian distribution and a Gaussian mixture model. If the random numbers generated by these two p.d.f.s have different probability distributions, the heterocluster is then split into smaller clusters. The second component merges the clusters that have high neighborhood densities into a homocluster, where the neighborhood density is measured using a novel evaluation criterion. In addition, a co‐association matrix is presented, which serves as a summary for the ensemble of diverse clusters. A series of experiments were conducted to evaluate the feasibility and effectiveness of the proposed ensemble member generation algorithm. Results show that the proposed MECE algorithm can select high quality ensemble members and as a result yield the better clusterings than six state‐of‐the‐art ensemble clustering algorithms, that is, cluster‐based similarity partitioning algorithm (CSPA), meta‐clustering algorithm (MCLA), hybrid bipartite graph formulation (HBGF), evidence accumulation clustering (EAC), locally weighted evidence accumulation (LWEA), and locally weighted graph partition (LWGP). Specifically, MECE algorithm has the nearly 23% higher average NMI, 27% higher average ARI, 15% higher average FMI, and 10% higher average purity than CSPA, MCLA, HBGF, EAC, LWEA, and LWGA algorithms. The experimental results demonstrate that MECE algorithm is a valid approach to deal with the clustering ensemble problems.

The synthetic minority oversampling technique (SMOTE) is a popular algorithm to reduce the impact of class imbalance in building classifiers, and has received several enhancements over the past 20 years. SMOTE and its variants synthesize a number of minority-class sample points in the original sample space to alleviate the adverse effects of class imbalance. This approach works well in many cases, but problems arise when synthetic sample points are generated in overlapping areas between different classes, which further complicates classifier training. To address this issue, this paper proposes a novel generalization-oriented rather than imputation-oriented minority-class sample point generation algorithm, named overlapping minimization SMOTE (OM-SMOTE). This algorithm is designed specifically for binary imbalanced classification problems. OM-SMOTE first maps the original sample points into a new sample space by balancing sample encoding and classifier generalization. Then, OM-SMOTE employs a set of sophisticated minority-class sample point imputation rules to generate synthetic sample points that are as far as possible from overlapping areas between classes. Extensive experiments have been conducted on 32 imbalanced datasets to validate the effectiveness of OM-SMOTE. Results show that using OM-SMOTE to generate synthetic minority-class sample points leads to better classifier training performances for the naive Bayes, support vector machine, decision tree, and logistic regression classifiers than the 11 state-of-the-art SMOTE-based imputation algorithms. This demonstrates that OM-SMOTE is a viable approach for supporting the training of high-quality classifiers for imbalanced classification. The implementation of OM-SMOTE is shared publicly on the GitHub platform at https://github.com/luxuan123123/OM-SMOTE/ .

Cutting-edge real-time detectors usually reach real-time performance by adopting lightweight architectures. The accuracy of detection may be limited by their insufficient capabilities to obtain powerful feature representation, which is a notoriously onerous task in machine vision applications. Aiming at this problem, this study proposes a method of adaptive aggregation of features at both scale and spatial levels in an anchor-free framework: 1) at the scale level, a Multi-scale Point Feature Fusion (MPFF) module has been proposed to fuse point features from multiple scales via a self-adaptive re-weighting manner; 2) at the spatial level, a Restrained Deformable Convolution (R-DCN) has been designed to focus on the most informative features in a pre-defined region while avoiding the remote feature distraction. Based on R-DCN, an Adaptive Spatial Aggregation (ASA) module has been presented to alleviate the feature misalignment problem in classification and regression tasks via their respective spatial divisions. Extensive experimental results on MS COCO indicate that Adaptive Aggregation Detector (AADet) achieves a state-of-the-art detection performance, i.e., 41.8 AP at 60 FPS.

Real-world sensors often produce incomplete, irregular, and noisy point clouds, making point cloud completion increasingly important. However, most existing completion methods rely on large paired datasets for training, which is labor-intensive. This paper proposes RaPD, a novel semi-supervised point cloud completion method that reduces the need for paired datasets. RaPD utilizes a two-stage training scheme, where a deep semantic prior is learned in stage 1 from unpaired complete and incomplete point clouds, and a semi-supervised prior distillation process is introduced in stage 2 to train a completion network using only a small number of paired samples. Additionally, a self-supervised completion module is introduced to improve performance using unpaired incomplete point clouds. Experiments on multiple datasets show that RaPD outperforms previous methods in both homologous and heterologous scenarios.

This paper proposes a novel and efficient method, called S-PDB, for the analysis and classification of Spike (S) protein structures of SARS-CoV-2 and other viruses/organisms in the Protein Data Bank (PDB). The method first finds and identifies protein structures in PDB that are similar to a protein structure of interest (SARS-CoV-2 S) via a protein structure comparison tool. The amino acid (AA) sequences of identified protein structures, downloaded from PDB, and their aligned amino acids (AAA) and secondary structure elements (ASSE), that are stored in three separate datasets, are then used for the reliable detection/classification of SARS-CoV-2 S protein structures. Three classifiers are used and their performance is compared by using six evaluation metrics. Obtained results show that two classifiers for text data (Multinomial Naive Bayes and Stochastic Gradient Descent) performed better and achieved high accuracy on the dataset that contains AAA of protein structures compared to the datasets for AA and ASSE, respectively.

Domain adaptation is common but challenging in signal processing tasks due to the intrinsic discrepancy, especially in difficult-to-label medical image segmentation application scenarios. Pseudo labeling methods are widely utilized to compensate for the scarcity of annotation. However, most existing methods set the fixed thresholds to select highly-confident predictions as pseudo labels, inevitably generating false labels with noise. In this paper, we combine the dual-classifiers consistency and predictive category-aware confidence to form a novel regularization for pseudo-label denoising. The dual-classifiers consistency helps promote the robustness of pseudo labels. Meanwhile, category-aware confidence is utilized as adaptive pixel-wise weights, avoiding the need for handcrafted thresholds. The adapted model is refined by the rectified pseudo labels without source domain samples. The proposed method is model-independent and thus can be plug-and-play to improve existing UDA methods. We validated it on the cross-modality medical image segmentation and obtained more competitive results.

Semi-Supervised Object Detection (SSOD) aims to improve performance by leveraging a large amount of unlabeled data. Existing works usually adopt the teacher-student framework to enforce student to learn consistent predictions over the pseudo-labels generated by teacher. However, the performance of the student model is limited since the noise inherently exists in pseudo-labels. In this paper, we investigate the causes and effects of noisy pseudo-labels and propose a simple yet effective approach denoted as Self-Correction Mean Teacher(SCMT) to reduce the adverse effects. Specifically, we propose to dynamically re-weight the unsupervised loss of each student's proposal with additional supervision information from the teacher model, and assign smaller loss weights to possible noisy proposals. Extensive experiments on MS-COCO benchmark have shown the superiority of our proposed SCMT, which can significantly improve the supervised baseline by more than 11% mAP under all 1%, 5% and 10% COCO-standard settings, and surpasses state-of-the-art methods by about 1.5% mAP. Even under the challenging COCO-additional setting, SCMT still improves the supervised baseline by 4.9% mAP, and significantly outperforms previous methods by 1.2% mAP, achieving a new state-of-the-art performance.

To alleviate the attribute independence assumption, this paper proposes an auto-encoding naive Bayesian classifier (AE-NBC) which is trained based on transformed independent condition-attributes using an auto-encoding strategy. First, an auto-encoding neural network (AENN) with single hidden-layer is created to map the original data set to a high-dimensional data space. To guarantee the independence among auto-encoding condition-attributes (AECAs), an efficient and convergent objective function is elaborated to optimize the input-layer and output-layer weights of the AENN by considering both the auto-encoding error and attribute independence degree. Second, independent component analysis is iteratively carried out to remove the redundant attributes in AECAs so that the optimal transformed condition-attributes (TCAs) are obtained for the training of the NBC. Experiments were conducted on 29 benchmark data sets to validate the rationality, feasibility, and effectiveness of AE- NBC. Experimental results show that the independence among TCAs is stronger than among the original condition-attributes, and that AE- NBC can obtain higher training and testing accuracies than six Bayesian classifiers.

This article presents a Bayesian attributebagging-based extreme learning machine (BAB-ELM)to handle high-dimensional classification and regression problems. First, thedecision-making degree (DMD)of a condition attribute is calculated based on the Bayesian decision theory, i.e., the conditional probability of the condition attribute given the decision attribute. Second, the condition attribute with the highest DMD is put into thecondition attribute group (CAG)corresponding to the specific decision attribute. Third, thebagging attribute groups (BAGs)are used to train an ensemble learning model ofextreme learning machines (ELMs).Each base ELM is trained on a BAG which is composed of condition attributes that are randomly selected from the CAGs. Fourth, the information amount ratios of bagging condition attributes to all condition attributes is used as the weights to fuse the predictions of base ELMs in BAB-ELM. Exhaustive experiments have been conducted to compare the feasibility and effectiveness of BAB-ELM with seven other ELM models, i.e., ELM, ensemble-based ELM (EN-ELM), voting-based ELM (V-ELM), ensemble ELM (E-ELM), ensemble ELM based on multi-activation functions (MAF-EELM), bagging ELM, and simple ensemble ELM. Experimental results show that BAB-ELM is convergent with the increase of base ELMs and also can yield higher classification accuracy and lower regression error for high-dimensional classification and regression problems.

Cross-modality adaptation is challenging due to the internal domain discrepancy in appearance and representation. When the trained model of source domain is transferred to the target domain, the domain shift will reduce accuracy. Meanwhile, unsupervised domain adaptation has the potential to recover this degradation among medical images of different modalities, so it is of clinical significance and meaningful in bioinformatics. However, previous related works usually try to align domains in a single direction or two directions, failing to take advantage of the complementary relationship between different directions and alignment tasks. In this paper, we propose the Tri-directional learning framework to solve domain shift in the task of medical image semantic segmentation. The proposed framework is able to synergize image style transformation, mask segmentation and edge segmentation. The above three tasks are mutually boosted through complementary training in each iteration. In this way, our method performs cross-modality medical images semantic segmentation from labeled source domain (MRI) to unlabeled target domain (CT). The experimental results demonstrate the effectiveness of the proposed method. For the task of cardiac structure segmentation from cross-modality medical images, our proposed framework achieves state-of-the-art performance. The code is available at https://github.com/lichen14/TriDL.

Due to the intrinsic domain shift among different modalities, it is nontrivial to directly apply a well-trained model into other cross-modality medical images. Unsupervised domain adaptation (UDA) has the potential to reduce such domain shift. However, existing UDA methods try to align domains in either image level or in feature level, failing to consider the unified relationship between cross-modality images and their corresponding features. In this paper, we propose a novel UDA framework for domain adaptive medical image segmentation. The proposed framework synergizes both pixel space and entropy space for domain alignment. Specifically, in the pixel space, we introduce the attention mechanism into CycleGAN, and enhance the semantic and geometric consistency of the target organs during the image style transformation. In entropy space, we utilize entropy minimization principle to force consistent image segmentation between well-annotated source domain and non-annotated target domain. The aligned ensemble of two representation spaces enables a well-trained segmentation model to effectively transfer from source domain to target domain. The experimental results demonstrate the effectiveness of the proposed method. For the task of multi-organs segmentation from cross-modality medical images, our proposed framework achieves state-of-the-art performance, with some specific metric even superior to those of supervised methods. The code is available at https://github.com/lichen14/AttENT.

A novel random sample partition-based clustering ensemble (RSP-CE) algorithm is proposed in this paper to handle the big data clustering problems. There are three key components in RSP-CE algorithm, i.e., generating the base clustering results on RSP data blocks, harmonizing the based clustering results with maximum mean discrepancy (MMD) criterion, and refining the RSP clustering results. RSP data blocks have the consistent sample distributions with the whole big data and thus provide the possibility for using base clustering results on different data subsets to approximate the clustering result on whole big data. The experimental results in comparison with other 5 well-known clustering ensemble algorithms on 4 big data sets show that RSP-CE algorithm obtains the better normalized mutual information (NMI) values and Fowlkes-Mallows Index (FMI) values with the less training time consumptions and thus demonstrate that RSP-CE algorithm is a viable approach to deal with the big data clustering problems.

Due to the ubiquitous presence of missing values in real-world datasets, an imputation algorithm can recover the missing values and provide users with a complete dataset that utilizes all the available observed information. However, most of the imputation methods still have several limitations, including cannot restore the original distribution, handling various data missing patterns, and high missing rate dataset. In this poster, a novel neural network-based two-stage missing value imputation (abbreviated as TS-MVI) method is proposed to fill an incomplete condition attribute with the optimized attribute values for the supervised learning task. By initializing the missing values with random numbers, the imputation values are iteratively adjusted based on the new updating rule by minimizing both the autoencoder-oriented objective function and neural network-based classification error. The persuasive experiments show that TS-MVl method significantly outperforms current state-of-the-art imputation methods and thus demonstrate TS-MVI is a viable approach to deal with the missing value imputation problem.

This paper proposes a compressed hidden naive Bayesian (C-HNB) classifier which is an improved version of hidden naive Bayesian (HNB) by compressing the Bayesian network structure and calculating the attribute correlation with maximal information coefficient (MIC). In C-HNB, we remodel the Bayesian network structure based on the chain rule of joint probability distribution so that the number of hidden parent nodes in C-HNB is smaller than the number of hidden parent nodes in HNB, which reduces the training complexity of Bayesian network. In addition, the attribute correlation in C-HNB is calculated with MIC rather than the mutual information, which makes the Bayesian network more stable because the calculation of mutual information is severely influenced by the discretization of continuous attribute. On the selected KEEL benchmark data sets, we compare the classification performances of C-HNB with HNB. The comparative results show that C-HNB can obtain the better prediction accuracy with the less training time in comparison with HNB and thus the experimental results demonstrate the effectiveness of C-HNB.

It is desirable to maintain both high accuracy and runtime efficiency in lane detection. State-of-the-art methods mainly address the efficiency problem by direct compression of high-dimensional features. These methods usually suffer from information loss and cannot achieve satisfactory accuracy performance. To ensure the diversity of features and subsequently maintain information as much as possible, we introduce multi-frequency analysis into lane detection. Specifically, we propose a multi-spectral feature compressor (MSFC) based on two-dimensional (2D) discrete cosine transform (DCT) to compress features while preserving diversity information. We group features and associate each group with an individual frequency component, which incurs only 1/7 overhead of one-dimensional convolution operation but preserves more information. Moreover, to further enhance the discriminability of features, we design a multi-spectral lane feature aggregator (MSFA) based on one-dimensional (1D) DCT to aggregate features from each lane according to their corresponding frequency components. The proposed method outperforms the state-of-the-art methods (including LaneATT and UFLD) on TuSimple, CULane, and LLAMAS benchmarks. For example, our method achieves 76.32% F1 at 237 FPS and 76.98% F1 at 164 FPS on CULane, which is 1.23% and 0.30% higher than LaneATT. Our code and models are available at https://github.com/harrylin-hyl/MSLD.

Object detection has long been an important issue in the discipline of scene understanding. Existing researches mainly focus on the object itself, ignoring its surrounding environment. In fact, the surrounding environment provides abundant information to help detectors classify and locate objects. This paper proposes CRPDet, viz. CenterRepp Detector, a framework for object detection. The main function of CRPDet is accomplished by the CenterRepp module, which takes into account the surrounding environment by predicting the distribution of the central representative points. CenterRepp converts labeled object frames into the mean and standard variance of the sampling points' distribution. This helps increase the receptive field of objects, breaking the limitation of object frames. CenterRepp defines a position-fixed center point with significant weights, avoiding to sample all points in the surroundings. Experiments on the COCO test-dev detection benchmark demonstrates that our proposed CRPDet has comparable performance with state-of-the-art detectors, achieving 39.4 mAP with 51 FPS tested under single size input.

Recently, deep learning has witnessed dramatic progress in the medical image analysis field. In the precise treatment of cancer immunotherapy, the quantitative analysis of PD-L1 immunohistochemistry is of great importance. It is quite common that pathologists manually quantify the cell nuclei. This process is very time-consuming and error-prone. In this paper, we describe the development of a platform for PD-L1 pathological image quantitative analysis using deep learning approaches. As point-level annotations can provide a rough estimate of the object locations and classifications, this platform adopts a point-level supervision model to classify, localize, and count the PD-L1 cells nuclei. Presently, this platform has achieved an accurate quantitative analysis of PD-L1 for two types of carcinoma, and it is deployed in one of the first-class hospitals in China.

With the ever-increasing volume of data, alternative strategies are required to divide big data into statistically consistent data blocks that can be used directly as representative samples of the entire data set in big data analysis. In this paper, we propose the Random Sample Partition (RSP) distributed data model to represent a big data set as a set of disjoint data blocks, called RSP blocks. Each RSP block has a probability distribution similar to that of the entire data set. RSP blocks can be used to estimate the statistical properties of the data and build predictive models without computing the entire data set. We demonstrate the implications of the RSP model on sampling from big data and introduce a new RSP-based method for approximate big data analysis which can be applied to different scenarios in the industry. This method significantly reduces the computational burden of big data and increases the productivity of data scientists.

In many application areas, data that is being generated and processed goes beyond the petabyte scale. Analyzing such an increasing massive volume of data faces computational, as well as, statistical challenges. In order to solve these challenges, distributed and parallel processing frameworks have been used for implementing scalable data analysis algorithms. Nevertheless, processing the whole big data set at one time may exceed the available computing resources and the time requirements for some applications. Thus, approximate approaches can be used to achieve asymptotic analysis results, especially when data analysis algorithms are amenable to an approximate result rather than an exact one. However, most approximation approaches require taking a random sample of the data which is a nontrivial task when working with big data sets. In this paper, we employ ensemble learning as an approach for asymptotic analysis using randomly selected subsets (i.e. data blocks) of a big data set. We propose an asymptotic ensemble learning framework which depends on block-based sampling rather than record-based sampling. In order to demonstrate the feasibility and performance of this framework, we present an empirical analysis on real data sets. In addition to the scalability advantage, the experimental results show that several blocks of a data set are enough to get approximately the same results as those from using the whole data set.

When traditional sample selection methods are used to compress large data sets, the computational complexity turns out to be very high and it is really time consuming. To avoid these shortcomings, we propose a new method to select samples based on non-stable cut points. With the basic characteristic of convex function that its extreme values occur at the endpoints of intervals, the method measures the extent of a sample being endpoints by labeling non-stable cut points. Then we can select the samples with higher endpoint extent, which can avoid calculating the distances between samples. This method aims to compress the data sets and improve the computational efficiency without affecting the classification accuracy. Experiments show that the proposed algorithm performs very well on the compression of data sets with higher imbalance degree. Meanwhile, the method is experimentally confirmed to have strong noise-resistance.

Similar to the ensemble learning for classification, regression ensemble also tries to improve the prediction accuracy through combining several “weak” estimators which are usually high-variance and thus unstable. In this paper, we propose a new scheme of fusing the weak Priestley-Chao Kernel Estimators (PCKEs) based on Choquet fuzzy integral, which differs from all the existing models of regressor fusion. The new scheme uses Choquet fuzzy integral to fuse several target outputs from different PCKEs, in which the optimal bandwidths are obtained with cross-validation criteria. The key of applying fuzzy integral to PCKE fusion is the determination of fuzzy measure. Considering the advantage of particle swarm optimization (PSO) algorithm on convergence rate, we use three different PSO algorithms, i.e., standard PSO (SPSO), Gaussian PSO (GPSO) and GPSO with Gaussian jump (GPSOGJ), to determine the general and λ fuzzy measures. The finally experimental results on 6 standard testing functions show that the new paradigm for regression ensemble based on fuzzy integral is more accurate and stable in comparison with any individual PCKE. This demonstrates the feasibility and effectiveness of our proposed regression ensemble model.

Variable Precision Rough Set (VPRS) model is one of the most important extensions of the Classical Rough Set (RS) theory. It employs a majority inclusion relation mechanism in order to make the Classical RS model become more fault tolerant, and therefore the generalization of the model is improved. This paper can be viewed as an extension of previous investigations on attribution reduction problem in VPRS model. In our investigation, we illustrated with examples that the previously proposed reduct definitions may spoil the hidden classification ability of a knowledge system by ignoring certian essential attributes in some circumstances. Consequently, by proposing a new ß-consistent notion, we analyze the relationship between the structures of Decision Table (DT) and different definitions of reduct in VPRS model. Then we give a new notion of /3-complement reduct that can avoid the defects of reduct notions defined in previous literatures. We also supply the method to obtain the /3- complement reduct using a decision table splitting algorithm, and finally demonstrate the feasibility of our approach with sample instances.

The objective of link prediction for social network is to estimate the likelihood that a link exists between two nodes x and y. There are some well-known local information-based link prediction algorithms (LILPAs) which have been proposed to handle this essential and crucial problem in the social network analysis. However, they can not adequately consider the so-called local information: the degrees of x and y, the number of common neighbors of nodes x and y, and the degrees of common neighbors of x and y. In other words, not any LILPA takes into account these local information simultaneously. This limits the performances of LILPAs to a certain degree and leads to the high variability of LILPAs. Thus, in order to make full use of all the local information and obtain a LILPA with highly-predicted capability, an ordered weighted averaging (OWA) operator based link prediction ensemble algorithm (LPEOWA) is proposed by integrating nine different LILPAs with aggregation weights which are determined with maximum entropy method. The final experimental results on benchmark social network datasets show that LPEOWA can obtain higher prediction accuracies which is measured by the area under the receiver operating characteristic curve (AUC) in comparison with nine individual LILPAs.

An important way to improve the performance of naive Bayesian classifiers (NBCs) is to remove or relax the fundamental assumption of independence among the attributes, which usually results in an estimation of joint probability density function (p.d.f.) instead of the estimation of marginal p.d.f. in the NBC design. This paper proposes a non-naive Bayesian classifier (NNBC) in which the independence assumption is removed and the marginal p.d.f. estimation is replaced by the joint p.d.f. estimation. A new technique of estimating the class-conditional p.d.f. based on the optimal bandwidth selection, which is the crucial part of the joint p.d.f. estimation, is applied in our NNBC. Three well-known indexes for measuring the performance of Bayesian classifiers, which are classification accuracy, area under receiver operating characteristic curve, and probability mean square error, are adopted to conduct a comparison among the four Bayesian models, i.e., normal naive Bayesian, flexible naive Bayesian (FNB), the homologous model of FNB (FNBROT), and our proposed NNBC. The comparative results show that NNBC is statistically superior to the other three models regarding the three indexes. And, in the comparison with support vector machine and four boosting-based classification methods, NNBC achieves a relatively favorable classification accuracy while significantly reducing the training time.

A new ontology learning model called domain ontology graph (DOG) is proposed in this paper. There are two key components in the DOG, i.e., the definition of the ontology graph and the ontology learning process. The former defines the ontology and knowledge conceptualization model from the domain-specific text documents; the latter offers the necessary method of semiautomatic domain ontology learning and generates the corresponding ontology graphs. Two kinds of ontological operations are also defined based on the proposed DOG, i.e., document ontology graph generation and ontology-graph-based text classification. The simulation studies focused upon Chinese text data are used to demonstrate the potential effectiveness of our proposed strategy. This is accomplished by generating DOGs to represent the domain knowledge and conducting the text classifications based on the generated ontology graph. The experimental results show that the new method can produce significantly better classification accuracy (e.g., with 92.3% in f-measure) compared with other methods (such as 86.8% in f-measure for the term-frequency-inverse-document-frequency approach). The high performance demonstrates that our presented ontological operations based on the ontology graph knowledge model are effectively developed.

In flexible Naive Bayesian (FNB), the excellent qualities of Gaussian kernel have been demonstrated by the theoretical analyses and experimental comparisons with normal Naive Bayesian(NNB). There are also several types of kernel functions commonly used for probability density estimation, i.e., uniform, triangular, epanechnikov, biweight, triweight and cosine. We call them discontinuous kernels. In this paper, we verify the feasibility and efficiency of applying these alternative kernels in FNB. Our works mainly focus on three aspects: firstly, we give the application conditions of these kernels for the given domain data by analyzing the structural difference between the discontinuous kernel and Gaussian kernel; secondly, the equivalent probability is proposed to improve the capabilities of discontinuous kernels when such problem of kernel incapability occurs; finally, we carry out the experimental demonstration of our proposed method based on 15 UCI datasets. The results show that the discontinuous kernels can obtain better classification accuracies with the help of equivalent probabilities.