Shanmin Pang

Xingyu Chen, Jianru Xue, Shanmin Pang

Vision-based localization in a prior map is of crucial importance for autonomous vehicles. Given a query image, the goal is to estimate the camera pose corresponding to the prior map, and the key is the registration problem of camera images within the map. While autonomous vehicles drive on the road under occlusion (e.g., car, bus, truck) and changing environment appearance (e.g., illumination changes, seasonal variation), existing approaches rely heavily on dense point descriptors at the feature level to solve the registration problem, entangling features with appearance and occlusion. As a result, they often fail to estimate the correct poses. To address these issues, we propose a sparse semantic map-based monocular localization method, which solves 2D-3D registration via a well-designed deep neural network. Given a sparse semantic map that consists of simplified elements (e.g., pole lines, traffic sign midpoints) with multiple semantic labels, the camera pose is then estimated by learning the corresponding features between the 2D semantic elements from the image and the 3D elements from the sparse semantic map. The proposed sparse semantic map-based localization approach is robust against occlusion and long-term appearance changes in the environments. Extensive experimental results show that the proposed method outperforms the state-of-the-art approaches.

Qirui Wang, Qi Guo, Yiding Sun et al.

Personalized text-to-image generation lets users fine-tune diffusion models into repositories of concept-specific checkpoints, but serving these repositories efficiently is difficult for two reasons: natural-language requests are often ambiguous and can be misrouted to visually similar checkpoints, and standard post-training quantization can distort the fragile representations that encode personalized concepts. We present PersonalQ, a unified framework that connects checkpoint selection and quantization through a shared signal -- the checkpoint's trigger token. Check-in performs intent-aligned selection by combining intent-aware hybrid retrieval with LLM-based reranking over checkpoint context and asks a brief clarification question only when multiple intents remain plausible; it then rewrites the prompt by inserting the selected checkpoint's canonical trigger. Complementing this, Trigger-Aware Quantization (TAQ) applies trigger-aware mixed precision in cross-attention, preserving trigger-conditioned key/value rows (and their attention weights) while aggressively quantizing the remaining pathways for memory-efficient inference. Experiments show that PersonalQ improves intent alignment over retrieval and reranking baselines, while TAQ consistently offers a stronger compression-quality trade-off than prior diffusion PTQ methods, enabling scalable serving of personalized checkpoints without sacrificing fidelity.

Lizhen Xu, Zehao Wu, Wenzhao Qiu et al.

Query-based models are extensively used in 3D object detection tasks, with a wide range of pre-trained checkpoints readily available online. However, despite their popularity, these models often require an excessive number of object queries, far surpassing the actual number of objects to detect. The redundant queries result in unnecessary computational and memory costs. In this paper, we find that not all queries contribute equally -- a significant portion of queries have a much smaller impact compared to others. Based on this observation, we propose an embarrassingly simple approach called Gradually Pruning Queries (GPQ), which prunes queries incrementally based on their classification scores. A key advantage of GPQ is that it requires no additional learnable parameters. It is straightforward to implement in any query-based method, as it can be seamlessly integrated as a fine-tuning step using an existing checkpoint after training. With GPQ, users can easily generate multiple models with fewer queries, starting from a checkpoint with an excessive number of queries. Experiments on various advanced 3D detectors show that GPQ effectively reduces redundant queries while maintaining performance. Using our method, model inference on desktop GPUs can be accelerated by up to 1.35x. Moreover, after deployment on edge devices, it achieves up to a 67.86% reduction in FLOPs and a 65.16% decrease in inference time. The code will be available at https://github.com/iseri27/Gpq.

Wenzhao Qiu, Shanmin Pang, Hao zhang et al.

Recent advances in high-definition (HD) map construction from surround-view images have highlighted their cost-effectiveness in deployment. However, prevailing techniques often fall short in accurately extracting and utilizing road features, as well as in the implementation of view transformation. In response, we introduce HeightMapNet, a novel framework that establishes a dynamic relationship between image features and road surface height distributions. By integrating height priors, our approach refines the accuracy of Bird's-Eye-View (BEV) features beyond conventional methods. HeightMapNet also introduces a foreground-background separation network that sharply distinguishes between critical road elements and extraneous background components, enabling precise focus on detailed road micro-features. Additionally, our method leverages multi-scale features within the BEV space, optimally utilizing spatial geometric information to boost model performance. HeightMapNet has shown exceptional results on the challenging nuScenes and Argoverse 2 datasets, outperforming several widely recognized approaches. The code will be available at \url{https://github.com/adasfag/HeightMapNet/}.

Qi Guo, Xiaojun Jia, Shanmin Pang et al.

Multimodal Large Language Models (MLLMs) are becoming integral to autonomous driving (AD) systems due to their strong vision-language reasoning capabilities. However, MLLMs are vulnerable to adversarial attacks, particularly adversarial patch attacks, which can pose serious threats in real-world scenarios. Existing patch-based attack methods are primarily designed for object detection models and perform poorly when transferred to MLLM-based systems due to the latter's complex architectures and reasoning abilities. To address these limitations, we propose PhysPatch, a physically realizable and transferable adversarial patch framework tailored for MLLM-based AD systems. PhysPatch jointly optimizes patch location, shape, and content to enhance attack effectiveness and real-world applicability. It introduces a semantic-based mask initialization strategy for realistic placement, an SVD-based local alignment loss with patch-guided crop-resize to improve transferability, and a potential field-based mask refinement method. Extensive experiments across open-source, commercial, and reasoning-capable MLLMs demonstrate that PhysPatch significantly outperforms prior methods in steering MLLM-based AD systems toward target-aligned perception and planning outputs. Moreover, PhysPatch consistently places adversarial patches in physically feasible regions of AD scenes, ensuring strong real-world applicability and deployability.

Lizhen Xu, Xiuxiu Bai, Xiaojun Jia et al.

Query-based methods with dense features have demonstrated remarkable success in 3D object detection tasks. However, the computational demands of these models, particularly with large image sizes and multiple transformer layers, pose significant challenges for efficient running on edge devices. Existing pruning and distillation methods either need retraining or are designed for ViT models, which are hard to migrate to 3D detectors. To address this issue, we propose a zero-shot runtime pruning method for transformer decoders in 3D object detection models. The method, termed tgGBC (trim keys gradually Guided By Classification scores), systematically trims keys in transformer modules based on their importance. We expand the classification score to multiply it with the attention map to get the importance score of each key and then prune certain keys after each transformer layer according to their importance scores. Our method achieves a 1.99x speedup in the transformer decoder of the latest ToC3D model, with only a minimal performance loss of less than 1%. Interestingly, for certain models, our method even enhances their performance. Moreover, we deploy 3D detectors with tgGBC on an edge device, further validating the effectiveness of our method. The code can be found at https://github.com/iseri27/tg_gbc.

Junkai Yang, Qirui Wang, Yaoqing Jin et al.

Retrieving partially relevant segments from untrimmed videos remains difficult due to two persistent challenges: the mismatch in information density between text and video segments, and limited attention mechanisms that overlook semantic focus and event correlations. We present KDC-Net, a Knowledge-Refined Dual Context-Aware Network that tackles these issues from both textual and visual perspectives. On the text side, a Hierarchical Semantic Aggregation module captures and adaptively fuses multi-scale phrase cues to enrich query semantics. On the video side, a Dynamic Temporal Attention mechanism employs relative positional encoding and adaptive temporal windows to highlight key events with local temporal coherence. Additionally, a dynamic CLIP-based distillation strategy, enhanced with temporal-continuity-aware refinement, ensures segment-aware and objective-aligned knowledge transfer. Experiments on PRVR benchmarks show that KDC-Net consistently outperforms state-of-the-art methods, especially under low moment-to-video ratios.

Qi Guo, Shanmin Pang, Xiaojun Jia et al.

Adversarial attacks, particularly \textbf{targeted} transfer-based attacks, can be used to assess the adversarial robustness of large visual-language models (VLMs), allowing for a more thorough examination of potential security flaws before deployment. However, previous transfer-based adversarial attacks incur high costs due to high iteration counts and complex method structure. Furthermore, due to the unnaturalness of adversarial semantics, the generated adversarial examples have low transferability. These issues limit the utility of existing methods for assessing robustness. To address these issues, we propose AdvDiffVLM, which uses diffusion models to generate natural, unrestricted and targeted adversarial examples via score matching. Specifically, AdvDiffVLM uses Adaptive Ensemble Gradient Estimation to modify the score during the diffusion model's reverse generation process, ensuring that the produced adversarial examples have natural adversarial targeted semantics, which improves their transferability. Simultaneously, to improve the quality of adversarial examples, we use the GradCAM-guided Mask method to disperse adversarial semantics throughout the image rather than concentrating them in a single area. Finally, AdvDiffVLM embeds more target semantics into adversarial examples after multiple iterations. Experimental results show that our method generates adversarial examples 5x to 10x faster than state-of-the-art transfer-based adversarial attacks while maintaining higher quality adversarial examples. Furthermore, compared to previous transfer-based adversarial attacks, the adversarial examples generated by our method have better transferability. Notably, AdvDiffVLM can successfully attack a variety of commercial VLMs in a black-box environment, including GPT-4V.

Lei-lei Li, Jianwu Fang, Junbin Xiao et al.

Egocentricly comprehending the causes and effects of car accidents is crucial for the safety of self-driving cars, and synthesizing causal-entity reflected accident videos can facilitate the capability test to respond to unaffordable accidents in reality. However, incorporating causal relations as seen in real-world videos into synthetic videos remains challenging. This work argues that precisely identifying the accident participants and capturing their related behaviors are of critical importance. In this regard, we propose a novel diffusion model, Causal-VidSyn, for synthesizing egocentric traffic accident videos. To enable causal entity grounding in video diffusion, Causal-VidSyn leverages the cause descriptions and driver fixations to identify the accident participants and behaviors, facilitated by accident reason answering and gaze-conditioned selection modules. To support Causal-VidSyn, we further construct Drive-Gaze, the largest driver gaze dataset (with 1.54M frames of fixations) in driving accident scenarios. Extensive experiments show that Causal-VidSyn surpasses state-of-the-art video diffusion models in terms of frame quality and causal sensitivity in various tasks, including accident video editing, normal-to-accident video diffusion, and text-to-video generation.

Tianyu Wei, Shanmin Pang, Qi Guo et al.

Text-to-image diffusion models can generate realistic images based on textual inputs, enabling users to convey their opinions visually through language. Meanwhile, within language, emotion plays a crucial role in expressing personal opinions in our daily lives and the inclusion of maliciously negative content can lead users astray, exacerbating negative emotions. Recognizing the success of diffusion models and the significance of emotion, we investigate a previously overlooked risk associated with text-to-image diffusion models, that is, utilizing emotion in the input texts to introduce negative content and provoke unfavorable emotions in users. Specifically, we identify a new backdoor attack, i.e., emotion-aware backdoor attack (EmoAttack), which introduces malicious negative content triggered by emotional texts during image generation. We formulate such an attack as a diffusion personalization problem to avoid extensive model retraining and propose the EmoBooth. Unlike existing personalization methods, our approach fine-tunes a pre-trained diffusion model by establishing a mapping between a cluster of emotional words and a given reference image containing malicious negative content. To validate the effectiveness of our method, we built a dataset and conducted extensive analysis and discussion about its effectiveness. Given consumers' widespread use of diffusion models, uncovering this threat is critical for society.

Zhikai Chen, Lingxi Xie, Shanmin Pang et al.

Deepfakes raised serious concerns on the authenticity of visual contents. Prior works revealed the possibility to disrupt deepfakes by adding adversarial perturbations to the source data, but we argue that the threat has not been eliminated yet. This paper presents MagDR, a mask-guided detection and reconstruction pipeline for defending deepfakes from adversarial attacks. MagDR starts with a detection module that defines a few criteria to judge the abnormality of the output of deepfakes, and then uses it to guide a learnable reconstruction procedure. Adaptive masks are extracted to capture the change in local facial regions. In experiments, MagDR defends three main tasks of deepfakes, and the learned reconstruction pipeline transfers across input data, showing promising performance in defending both black-box and white-box attacks.

Qinghai Zheng, Jihua Zhu, Zhongyu Li et al.

This study investigates the problem of multi-view clustering, where multiple views contain consistent information and each view also includes complementary information. Exploration of all information is crucial for good multi-view clustering. However, most traditional methods blindly or crudely combine multiple views for clustering and are unable to fully exploit the valuable information. Therefore, we propose a method that involves consistent and complementary graph-regularized multi-view subspace clustering (GRMSC), which simultaneously integrates a consistent graph regularizer with a complementary graph regularizer into the objective function. In particular, the consistent graph regularizer learns the intrinsic affinity relationship of data points shared by all views. The complementary graph regularizer investigates the specific information of multiple views. It is noteworthy that the consistent and complementary regularizers are formulated by two different graphs constructed from the first-order proximity and second-order proximity of multiple views, respectively. The objective function is optimized by the augmented Lagrangian multiplier method in order to achieve multi-view clustering. Extensive experiments on six benchmark datasets serve to validate the effectiveness of the proposed method over other state-of-the-art multi-view clustering methods.

Xinsheng Wang, Shanmin Pang, Jihua Zhu

In the generalized zero-shot learning, synthesizing unseen data with generative models has been the most popular method to address the imbalance of training data between seen and unseen classes. However, this method requires that the unseen semantic information is available during the training stage, and training generative models is not trivial. Given that the generator of these models can only be trained with seen classes, we argue that synthesizing unseen data may not be an ideal approach for addressing the domain shift caused by the imbalance of the training data. In this paper, we propose to realize the generalized zero-shot recognition in different domains. Thus, unseen (seen) classes can avoid the effect of the seen (unseen) classes. In practice, we propose a threshold and probabilistic distribution joint method to segment the testing instances into seen, unseen and uncertain domains. Moreover, the uncertain domain is further adjusted to alleviate the domain shift. Extensive experiments on five benchmark datasets show that the proposed method exhibits competitive performance compared with that based on generative models.

Zhikai Chen, Lingxi Xie, Shanmin Pang et al.

There have been many efforts in attacking image classification models with adversarial perturbations, but the same topic on video classification has not yet been thoroughly studied. This paper presents a novel idea of video-based attack, which appends a few dummy frames (e.g., containing the texts of `thanks for watching') to a video clip and then adds adversarial perturbations only on these new frames. Our approach enjoys three major benefits, namely, a high success rate, a low perceptibility, and a strong ability in transferring across different networks. These benefits mostly come from the common dummy frame which pushes all samples towards the boundary of classification. On the other hand, such attacks are easily to be concealed since most people would not notice the abnormality behind the perturbed video clips. We perform experiments on two popular datasets with six state-of-the-art video classification models, and demonstrate the effectiveness of our approach in the scenario of universal video attacks.

Xinsheng Wang, Shanmin Pang, Jihua Zhu et al.

Zero-shot learning, which aims to recognize new categories that are not included in the training set, has gained popularity owing to its potential ability in the real-word applications. Zero-shot learning models rely on learning an embedding space, where both semantic descriptions of classes and visual features of instances can be embedded for nearest neighbor search. Recently, most of the existing works consider the visual space formulated by deep visual features as an ideal choice of the embedding space. However, the discrete distribution of instances in the visual space makes the data structure unremarkable. We argue that optimizing the visual space is crucial as it allows semantic vectors to be embedded into the visual space more effectively. In this work, we propose two strategies to accomplish this purpose. One is the visual prototype based method, which learns a visual prototype for each visual class, so that, in the visual space, a class can be represented by a prototype feature instead of a series of discrete visual features. The other is to optimize the visual feature structure in an intermediate embedding space, and in this method we successfully devise a multilayer perceptron framework based algorithm that is able to learn the common intermediate embedding space and meanwhile to make the visual data structure more distinctive. Through extensive experimental evaluation on four benchmark datasets, we demonstrate that optimizing visual space is beneficial for zero-shot learning. Besides, the proposed prototype based method achieves the new state-of-the-art performance.

Qinghai Zheng, Jihua Zhu, Zhiqiang Tian et al.

Multi-view clustering is an important and fundamental problem. Many multi-view subspace clustering methods have been proposed, and most of them assume that all views share a same coefficient matrix. However, the underlying information of multi-view data are not fully exploited under this assumption, since the coefficient matrices of different views should have the same clustering properties rather than be uniform among multiple views. To this end, this paper proposes a novel Constrained Bilinear Factorization Multi-view Subspace Clustering (CBF-MSC) method. Specifically, the bilinear factorization with an orthonormality constraint and a low-rank constraint is imposed for all coefficient matrices to make them have the same trace-norm instead of being equivalent, so as to explore the consensus information of multi-view data more fully. Finally, an Augmented Lagrangian Multiplier (ALM) based algorithm is designed to optimize the objective function. Comprehensive experiments tested on nine benchmark datasets validate the effectiveness and competitiveness of the proposed approach compared with several state-of-the-arts.

Qinghai Zheng, Jihua Zhu, Zhongyu Li et al.

Multi-view clustering is a learning paradigm based on multi-view data. Since statistic properties of different views are diverse, even incompatible, few approaches implement multi-view clustering based on the concatenated features straightforward. However, feature concatenation is a natural way to combine multi-view data. To this end, this paper proposes a novel multi-view subspace clustering approach dubbed Feature Concatenation Multi-view Subspace Clustering (FCMSC), which boosts the clustering performance by exploring the consensus information of multi-view data. Specifically, multi-view data are concatenated into a joint representation firstly, then, $l_{2,1}$-norm is integrated into the objective function to deal with the sample-specific and cluster-specific corruptions of multiple views. Moreover, a graph regularized FCMSC is also proposed in this paper to explore both the consensus information and complementary information of multi-view data for clustering. It is noteworthy that the obtained coefficient matrix is not derived by simply applying the Low-Rank Representation (LRR) to concatenated features directly. Finally, an effective algorithm based on the Augmented Lagrangian Multiplier (ALM) is designed to optimize the objective functions. Comprehensive experiments on six real-world datasets illustrate the superiority of the proposed methods over several state-of-the-art approaches for multi-view clustering.

Shanmin Pang, Jin Ma, Jianru Xue et al.

Image retrieval based on deep convolutional features has demonstrated state-of-the-art performance in popular benchmarks. In this paper, we present a unified solution to address deep convolutional feature aggregation and image re-ranking by simulating the dynamics of heat diffusion. A distinctive problem in image retrieval is that repetitive or \emph{bursty} features tend to dominate final image representations, resulting in representations less distinguishable. We show that by considering each deep feature as a heat source, our unsupervised aggregation method is able to avoid over-representation of \emph{bursty} features. We additionally provide a practical solution for the proposed aggregation method and further show the efficiency of our method in experimental evaluation. Inspired by the aforementioned deep feature aggregation method, we also propose a method to re-rank a number of top ranked images for a given query image by considering the query as the heat source. Finally, we extensively evaluate the proposed approach with pre-trained and fine-tuned deep networks on common public benchmarks and show superior performance compared to previous work.

Jihua Zhu, Siyu Xu, Zutao Jiang et al.

This paper proposes a global approach for the multi-view registration of unordered range scans. As the basis of multi-view registration, pair-wise registration is very pivotal. Therefore, we first select a good descriptor and accelerate its correspondence propagation for the pair-wise registration. Then, we design an effective rule to judge the reliability of pair-wise registration results. Subsequently, we propose a model augmentation method, which can utilize reliable results of pair-wise registration to augment the model shape. Finally, multi-view registration can be accomplished by operating the pair-wise registration and judgment, and model augmentation alternately. Experimental results on public available data sets show, that this approach can automatically achieve the multi-view registration of unordered range scans with good accuracy and effectiveness.

Jiaxing Wang, Jihua Zhu, Shanmin Pang et al.

Aggregating deep convolutional features into a global image vector has attracted sustained attention in image retrieval. In this paper, we propose an efficient unsupervised aggregation method that uses an adaptive Gaussian filter and an elementvalue sensitive vector to co-weight deep features. Specifically, the Gaussian filter assigns large weights to features of region-of-interests (RoI) by adaptively determining the RoI's center, while the element-value sensitive channel vector suppresses burstiness phenomenon by assigning small weights to feature maps with large sum values of all locations. Experimental results on benchmark datasets validate the proposed two weighting schemes both effectively improve the discrimination power of image vectors. Furthermore, with the same experimental setting, our method outperforms other very recent aggregation approaches by a considerable margin.

Zutao Jiang, Jihua Zhu, Georgios D. Evangelidis et al.

Generally, there are three main factors that determine the practical usability of registration, i.e., accuracy, robustness, and efficiency. In real-time applications, efficiency and robustness are more important. To promote these two abilities, we cast the multi-view registration into a clustering task. All the centroids are uniformly sampled from the initially aligned point sets involved in the multi-view registration, which makes it rather efficient and effective for the clustering. Then, each point is assigned to a single cluster and each cluster centroid is updated accordingly. Subsequently, the shape comprised by all cluster centroids is used to sequentially estimate the rigid transformation for each point set. For accuracy and stability, clustering and transformation estimation are alternately and iteratively applied to all point sets. We tested our proposed approach on several benchmark datasets and compared it with state-of-the-art approaches. Experimental results validate its efficiency and robustness for the registration of multi-view point sets.

Congcong Jin, Jihua Zhu, Yaochen Li et al.

Recently, the low rank and sparse (LRS) matrix decomposition has been introduced as an effective mean to solve the multi-view registration. It views each available relative motion as a block element to reconstruct one matrix so as to approximate the low rank matrix, where global motions can be recovered for multi-view registration. However, this approach is sensitive to the sparsity of the reconstructed matrix and it treats all block elements equally in spite of their varied reliability. Therefore, this paper proposes an effective approach for multi-view registration by the weighted LRS decomposition. Based on the anti-symmetry property of relative motions, it firstly proposes a completion strategy to reduce the sparsity of the reconstructed matrix. The reduced sparsity of reconstructed matrix can improve the robustness of LRS decomposition. Then, it proposes the weighted LRS decomposition, where each block element is assigned with one estimated weight to denote its reliability. By introducing the weight, more accurate registration results can be recovered from the estimated low rank matrix with good efficiency. Experimental results tested on public data sets illustrate the superiority of the proposed approach over the state-of-the-art approaches on robustness, accuracy, and efficiency.