Large language models (LLMs) are often assumed to contain "safety regions” - parameter subsets whose modification directly influences safety behaviors. We conduct a systematic evaluation of four safety region identification methods spanning different parameter granularities, from individual weights to entire Transformer layers, across four families of backbone LLMs with varying sizes. Using ten safety identification datasets, we find that the identified safety regions exhibit only low to moderate overlap, as measured by IoU. The overlap drops significantly when the safety regions are further refined using utility datasets (i.e. non-harmful queries). These results suggest that current techniques fail to reliably identify a stable, dataset-agnostic safety region.

Multi‑page Document Visual Question Answering (MP‑DocVQA) remains challenging because long documents not only strain computational resources but also reduce the effectiveness of the attention mechanism in large vision–language models (LVLMs). We tackle these issues with an Adaptive Visual In‑document Retrieval (AVIR) framework. A lightweight retrieval model first scores each page for question relevance. Pages are then clustered according to the score distribution to adaptively select relevant content. The clustered pages are screened again by Top-K to keep the context compact. However, for short documents, clustering reliability decreases, so we use a relevance probability threshold to select pages. The selected pages alone are fed to a frozen LVLM for answer generation, eliminating the need for model fine‑tuning. The proposed AVIR framework reduces the average page count required for question answering by 70%, while achieving an ANLS of 84.58% on the MP-DocVQA dataset—surpassing previous methods with significantly lower computational cost. The effectiveness of the proposed AVIR is also verified on the SlideVQA and DUDE benchmarks. Our code will be made publicly available upon acceptance.

Transformer-based models have made significant progress in edge detection, but their high computational cost is prohibitive. Recently, vision Mamba have shown excellent ability in efficiently capturing long-range dependencies. Drawing inspiration from this, we propose a novel edge detector with Mamba, termed EDMB, to efficiently generate high-quality multi-granularity edges. In EDMB, Mamba is combined with a global-local architecture, therefore it can focus on both global information and fine-grained cues. The fine-grained cues play a crucial role in edge detection, but are usually ignored by ordinary Mamba. We design a novel decoder to construct learnable Gaussian distributions by fusing global features and fine-grained features. And the multi-grained edges are generated by sampling from the distributions. In order to make multi-granularity edges applicable to single-label data, we introduce Evidence Lower Bound loss to supervise the learning of the distributions. On the multi-label dataset BSDS500, our proposed EDMB achieves competitive single-granularity ODS 0.837 and multi-granularity ODS 0.851 without multi-scale test or extra PASCAL-VOC data. Remarkably, EDMB can be extended to single-label datasets such as NYUDv2 and BIPED. The source code is available at https://github.com/Li-yachuan/EDMB.

A crucial aspect in abusive language on social media platforms (toxicity, hate speech, harmful stereotypes, etc.) is its inherent contextual nature. In this paper, we focus on the role of conversational context in abusive language detection, one of the most “direct” forms of context in this domain, as given by the conversation threads (e.g., directly preceding message, original post). The incorporation of surrounding messages has proven vital for the accurate human annotation of harmful content. However, many prior works have either ignored this aspect, collecting and processing messages in isolation, or have obtained inconsistent results when attempting to embed such contextual information into traditional classification methods. The reasons behind these findings have not yet been properly addressed. To this end, we propose an analysis of the impact of conversational context in abusive language detection, through: (1) an analysis of prior works and the limitations of the most common concatenation-based approach, which we attempt to address with two alternative architectures; (2) an evaluation of these methods on existing datasets in English, and a new dataset of French tweets annotated for hate speech and stereotypes; and (3) a qualitative analysis showcasing the necessity for context-awareness in ALD, but also its difficulties.

Large deformation is a key issue for deformable medical image registration. Decomposing a large deformation into several small deformations is an efficient solution. Current decomposition methods can be classed into iteration-based and non-iterative-based approaches. However, compared to iteration-based methods, non-iterative-based methods have faster inference times but registration accuracy gaps. To alleviate this limitation, we design a novel Non-iterative Pyramid Network (NIPNet). Firstly, Our Dual-domain Feature Extraction Module (DFEM) extracts global and local features in the frequency and spatial domains, respectively. Hence, the model considers global and local deformations. Secondly, a Multi-scale Localization Information Fusion Module (MIFM) is applied to fuse the localization information of adjacent scales to assist the current level in obtaining a more accurate deformation field. Finally, a Pyramid Self-distillation Loss (PDL) is introduced to improve the registration accuracy by treating the final deformation field as a teacher to guide the intermediate deformation field. By conducting intensive experiments on two typical 3D brain MRI datasets, we verify that the proposed NIPNet outperforms SOTA iterative-based methods and requires only a similar runtime as non-iterative approaches. Code is available at https://github.com/JXT210/NIPNet.

Semi-supervised learning has been widely used to train models by utilizing a small amount of labeled data and a large amount of unlabeled data, especially in medical image analysis. However, existing methods still do not fully reveal the complex regions (e.g., small branches or fuzzy edges). These complex regions contain more critical information, and by focusing more on complex region information, the predictions of model can be more comprehensive. To address the above problems, we propose a new voxel mask-based cyclic consistency network for semi-supervised medical image segmentation. First, our model consists of two identical encoders and three different decoders. Learning from the masked image, the two encoders enhance perceptual analysis and learning of details while retaining high-level semantic knowledge. Second, a sharpening function transforms the probabilistic results into soft pseudo labels and the cyclic consistency regularization training strategy achieves a consistency constraint between the probabilistic outputs of the different decoders and the soft pseudo labels. We evaluate our network on two public 3D medical datasets. Experiments demonstrate the superior performance of our proposed model over other methods. Our code is available here.

Smoothing of the graph convolution is not conducive to characterizing local differences of point cloud. To solve this problem, we propose a Differential Graph Convolutional Network (Differ-GCN) for point cloud analysis. First, we propose a new graph construction strategy that can make similar nodes in the local space belong to the same graph, which can better represent the local commonality. After that, the features of the graph are extracted by the similarity matrix. Some of the smoothing information of the graph is removed to optimize the over-smoothing nodes and combined with the local difference of the points to get the beneficial features for downstream tasks. Finally, each neighbor point is processed to generate a mask, and pooling is performed through the mask to reduce information loss. The experiment results show that Differ-GCN performs excellent in object classification and part segmentation. The processing speed of Differ-GCN for point cloud is much faster than the state-of-the-art methods.

Abstract Environmental pollution is one of the major challenges in China, which seriously affects people's life, economy and society. This paper proposes a multi‐criteria group decision‐making method based on hesitant fuzzy preference relations with self‐confidence (HFPRs‐SC) behaviours and applies it to the evaluation of the environmental pollution emergency response process. Firstly, a comprehensive evaluation system is proposed that covers the environmental pollution emergency response process. Moreover, is utilized to obtain the normalized hesitant fuzzy preference relationship, and an algorithm that considers both the hesitant fuzzy preference value and the self‐confidence level is proposed to improve the consistency of HFPRs‐SC. In addition, combined with subjective and objective weights of experts, a weighted average operator based on the confidence level is applied to aggregate individual HFPR‐SC. Furthermore, the score function of HFPRs‐SC is designed to get the best ranking. Finally, a case study on an explosion accident in Fujian Province of China is conducted to verify the practicability and effectiveness of the proposed method with comparative analysis and sensitivity analysis. Based on the analysis, some suggestions are put forward for improving environmental pollution emergency response.

With the rapid development of autonomous driving, point cloud is becoming a hot research hotspot Although the current network has solved the irregular structures and sparsity of point clouds, the geometric transformation of point clouds remains challenging problems. We propose a new graph convolution method, which has the ability to learn the geometric invariance features of point clouds. We define the convolution kernel as a graph, and add the maximum angle feature between each edge of the convolution kernel and the receptive field graph. We propose a new graph pyramid pooling, which integrates multi-scale features to learn local features better. We evaluate our method on multiple point cloud analysis tasks, including shape classification and part segmentation. Experimental results show that our method presents the state-of-the-art performance on the rotation-augmented object classification and segmentation.

With the rapid development of automatic driving, point cloud recognition is becoming an important problem for 3D understanding. Among existing works, they don't take full advantage of local structural information and the relationship between points is difficult to be represented. In this paper, we propose graph attention neural network (GANN) to enhance the structural representation of point clouds by relational degree between points. We construct graph structure on point cloud and use the attention mechanism to compute the relationship of different nodes in a neighborhood, and then embedding them into the aggregation operation designed. Finally, node's features can be updated by multiple graph attention layer over time, so that we can obtain more plentiful node feature representation and local structural feature. Extensive experiments on popular benchmarks demonstrate that our methods are effective for point cloud recognition.

The applications of isometric 3-D objects have recently received sufficient attention and, thus, it is very attractive to retrieve such isometric 3-D objects from large-scale collections. Although existing approaches have presented some interesting ideas, their performance is limited to their ability on feature representation. To improve the performance of 3-D object (shape) recognition, some recent algorithms prefer using complicated deep neural networks to learn discriminative features, but they consume huge amounts of computing resources. Instead, this paper presents a more effective solution by seamlessly integrating the traditional local descriptor with a deep pointwise convolutional network to extract 1-D features for shape recognition and retrieval. To reduce the costs of designing a complicated deep network, the first step of our algorithm is to describe the shape deformation by sampling a set of intrinsic point descriptors. Then, we introduce a simple yet effective pointwise convolutional network to integrate these descriptors as a global feature and the learning process can be significantly accelerated with the help of downsampling. Furthermore, a knowledge transfer strategy is used to upgrade our feature by compensating for information loss. Finally, we carry out experimental evaluations over popular shape benchmarks, and the results suggest that our approach exhibits superior accuracy rates and robustness on shape recognition and retrieval.

In this paper, we demonstrate a practical system for automatic weather-oriented clothing suggestion, given the weather information, the system can automatically recommend the most suitable clothing from the user s personal clothing album, or intelligently suggest the most pairing one with the user-specified reference clothing. This is an extremely challenging problem due to the large discrepancy factors that should be considered under different weather conditions. To approach this task, we use clothing attributes as a mid-level bridge to narrow the gap between low-level features and the high-level weather categories. We adopt a scoring function, which includes three terms, to model the relationship. To acquire an optimized model and verify our proposed method, we collect a large clothing Weather-to-Garment (WoG) dataset. Experiments on the WoG dataset demonstrate that our learned model are effective for both weather-oriented clothing recommendation and pairing.

In this paper, we propose a new approach for cross-scenario clothing retrieval and fine-grained clothing style recognition. The query clothing photos captured by cameras or other mobile devices are filled with noisy background while the product clothing images online for shopping are usually presented in a pure environment. We tackle this problem by two steps. Firstly, a hierarchical super-pixel merging algorithm based on semantic segmentation is proposed to obtain the intact query clothing item. Secondly, aiming at solving the problem of clothing style recognition in different scenarios, we propose sparse coding based on domain-adaptive dictionary learning to improve the accuracy of the classifier and adaptability of the dictionary. In this way, we obtain fine-grained attributes of the clothing items and use the attributes matching score to re-rank the retrieval results further. The experiment results show that our method outperforms the state-of-the-art approaches. Furthermore, we build a well labeled clothing dataset, where the images are selected from 1.5 billion product clothing images.

Nowadays, the Bag-of-words (BoW) representation is well applied to recent state-of-the-art image retrieval works. However, with the rapid growth in the number of images, the dimension of the dictionary increases substantially which leads to great storage and CPU cost. Besides, the local features do not convey any semantic information which is very important in image retrieval. In this paper, we propose to use "topics" instead of "visual words" as the image representation by topic model to reduce the feature dimension and mine more high level semantic information. We call this as Bag-of-Topics (BoT) which is a type of statistical model for discovering the abstract "topics" from the words. We extract the topics by Latent Dirichlet Allocation (LDA) and calculate the similarity between the images using BoT model instead of BoW directly. The results show that the dimension of the image representation has been reduced significantly, while the retrieval performance is improved.

The manifold of the dataset turns out to be quite useful in refining the retrieval results, and the diffusion process provides an efficient solution by careful selection of the similarity neighborhood which is usually modeled as the K-nearest neighborhood (KNN) graph. However, existing works are sensitive to the topology noises induced by the first K neighbors. In this paper, we tackle the problem by studying metric transformation which aims at finding new functional relationship to dig the latent similarity. The advantage of the approach lies in its robustness towards the varying K values; that is to say, it could preserve high similarity performances even if K is very large. Except for discussing only the global KNN (i.e. the same K for all neighborhoods) graph, we also investigate to specify a different K for each neighborhood by incorporating the new penalized consensus information (PCI). We show that PCI works superior compared with the original consensus information for denoising. Experiments on multiple affinity matrices have corroborated the superiority of our method with surprising good results.

Deformable non-rigid 3D shape retrieval plays an important role in various applications. Although there are many related works, their precision and robustness are not ideal. In this paper, we develop a novel retrieval method by using graph contexts, which consists of three steps. Initially, we evaluate the performance of spectral distances for deformable shape representation, which has not been studied in detail before. Then, we create a weighted L2distance for similarity measurement based on the spectra of Laplace-Beltrami operator. Finally, a new local graph diffusion method is introduced to reduce the mismatch error in feature space and the time cost of diffusion has reduced a lot simultaneously. Our experiment results on SHREC'11 Non-rigid dataset have reached the best reported retrieval performance (MAP: 99.9%).

Deformable shape retrieval has posed a challenge to researchers, which is becoming more and more important. This paper addresses non-rigid 3D shape retrieval in terms of a subtle adjustment strategy. For this purpose, we first construct a new global shape descriptor based on biharmonic distance, which uses the connectivity between vertices for natural shape representation. Then, in order to enhance retrieval ability, we propose a subtle adjustment method to provide discriminative information. Finally, we depend on experiment to evaluate the performance of the proposed method. Comparison results have proved that the proposed GBI-SA retrieval method is superior to state-of-the-art methods, including SD-GDM, mesh SIFT, MDS-CM-BOF, Shape DNA, and WESD.

In this paper, we have developed a novel algorithm for book image retrieval that can find test book images accurately and efficiently. In our proposed algorithm, SIFT Vocabulary tree is first used to model the local information and to get the preliminary results. The final result is determined via multi-features joint decision. The multi-features are SIFT density, SIFT distribution histogram, edge histogram and Color Histogram. Experiments on a 30000 dataset have demonstrated the efficiency of our proposed method.

An efficient randomized algorithm for line and circle detection in the Conformal Geometric Algebra framework is designed and implemented. This arithmetic performs well when an image has strong background noise. In conformal geometric algebra, circle and line in conformal space can be represented as multivectors. Then the distance would be more easily computed from each edge point to the hypothesis by this detecting algorithm. Based on these distances, the matched points are efficiently summarized and aggregated to yield the true object. The experimental evaluation achieves excellent performance in the hypotheses voting and yields competitive results.

This paper proposes an extended 3-D shape retrieval feature, which is based on a new scheme. In our method, continuous principal component analysis is used to align 3-D models into a canonical position. We then represent 3-D polygonal mesh as a 3-D multi-layer closed curve. It can capture the more feature of the 3-D models. The feature of the 3-D model is extracted from the 3-D closed curve by the Fourier transform. And the experiments show it is more efficient.

With the development of computer techniques, 3D model has been used more and more widely and content-based 3D model retrieval has been a hotspot in the area of multimedia information retrieval. How to extract 3D models' feature effectively is still a difficulty. Projection based 3D model feature extraction is an important kind in feature extraction, because of its robustness against noise, simplification and its simplicity of feature extraction. Based on the theory of principle plane, a feature extraction method based on the projection of principle plane is proposed and applied to 3D model retrieval. The experiments show that the method performs well.

In this paper, we have developed a novel framework calledJustClickto enable personalized image recommendation via exploratory search from large-scale collections of Flickr images. First, a topic network is automatically generated to summarize large-scale collections of Flickr images at a semantic level. Hyperbolic visualization is further used to enable interactive navigation and exploration of the topic network, so that users can gain insights of large-scale image collections at the first glance, build up their mental query models interactively and specify their queries (i.e., image needs) more precisely by selecting the image topics on the topic network directly. Thus, our personalized query recommendation framework can effectively address both the problem of query formulation and the problem of vocabulary discrepancy and null returns. Second, a small set of most representative images are recommended for the given image topic according to their representativeness scores. Kernel principal component analysis and hyperbolic visualization are seamlessly integrated to organize and layout the recommended images (i.e., most representative images) according to their nonlinear visual similarity contexts, so that users can assess the relevance between the recommended images and their real query intentions interactively. An interactive interface is implemented to allow users to express their time-varying query intentions precisely and to direct ourJustClicksystem to more relevant images according to their personal preferences. Our experiments on large-scale collections of Flickr images show very positive results.

In this paper, a forward neural network (FNN) is used for 3D model retrieval. Also the descriptor based on exponentially decaying Euclidean distance transforms (EDT) is adapted to represent the feature of a 3D model. As a kind of machine learning method, FNN is trained by the PSB trained data, and then used to sort the testing data set in this contest.

A novel methodology based on V-system polynomials for content-based search and retrieval of 3D objects is proposed and implemented in the experiment with the Contest’s query set and the model database. By the benchmark of the SHREC’06 to the algorithm, this method shows highly performance and improves the retrieval efficiency.

Analysis of transforming matrices between Bezier basis functions and geometrically continuous basis functions is presented. It is shown that G/sup 2/ transforming matrix has some relationship with G/sup 1/ transforming matrix. Based on this relationship a method of calculating G/sup 1/ basis functions is introduced, the degree of which is higher than quadratic. These basis functions are expressed explicitly via matrices decomposition. Equations for constructing G/sup 2/ splines can be presented independently of geometric shape parameters' values.

This correspondence addresses the problem of rotation, scale, and translation invariant recognition of images. Object shape comparability is a challenging problem in the field of pattern recognition and computer vision. The method based on the geometric moment invariants is the typical method in these applications. This paper introduces curve structure moment invariants based on the geometric moment invariants from transforming the density in geometric moments into a new density. The difference in the shapes is increased by using the curve structure moment invariants. Therefore, this method can be used in object shape analysis. To support our new theory, an algorithm for object shape analysis is designed and experiments based on square transform are conducted. Experiments give an encouraging high recognition rate by using the curve structure moment invariants.