Multimodal Large Language Models (MLLMs) bridge the gap between visual and textual data, enabling a range of advanced applications. However, complex internal interactions among visual elements and their alignment with text can introduce vulnerabilities, which may be exploited to bypass safety mechanisms. To address this, we analyze the relationship between image content and task and find that the complexity of subimages, rather than their content, is key. Building on this insight, we propose the Distraction Hypothesis, followed by a novel framework called Contrasting Subimage Distraction Jailbreaking (CS-DJ), to achieve jailbreaking by disrupting MLLMs alignment through multi-level distraction strategies. CS-DJ consists of two components: structured distraction, achieved through query decomposition that induces a distributional shift by fragmenting harmful prompts into sub-queries, and visual-enhanced distraction, realized by constructing contrasting subimages to disrupt the interactions among visual elements within the model. This dual strategy disperses the model's attention, reducing its ability to detect and mitigate harmful content. Extensive experiments across five representative scenarios and four popular closed-source MLLMs, including \texttt GPT-4o-mini , \texttt GPT-4o , \texttt GPT-4V , and \texttt Gemini-1.5-Flash , demonstrate that CS-DJ achieves average success rates of 52.40% for the attack success rate and 74.10% for the ensemble attack success rate. These results reveal the potential of distraction-based approaches to exploit and bypass MLLMs' defenses, offering new insights for attack strategies.
论文检索
输入标题、作者或关键词，从 2,871 篇学术成果中精准定位
RoboPEPP: Vision-Based Robot Pose and Joint Angle Estimation through Embedding Predictive Pre-Training
PDF ↗Vision-based pose estimation of articulated robots with unknown joint angles has applications in collaborative robotics and human-robot interaction tasks. Current frameworks use neural network encoders to extract image features and downstream layers to predict joint angles and robot pose. While images of robots inherently contain rich information about the robot's physical structures, existing methods often fail to leverage it fully; therefore, limiting performance under occlusions and truncations. To address this, we introduce RoboPEPP, a method that fuses information about the robot's physical model into the encoder using a masking-based self-supervised embedding-predictive architecture. Specifically, we mask the robot's joints and pre-train an encoder-predictor model to infer the joints' embeddings from surrounding unmasked regions, enhancing the encoder's understanding of the robot's physical model. The pre-trained encoder-predictor pair, along with joint angle and keypoint prediction networks, is then fine-tuned for pose and joint angle estimation. Random masking of input during fine-tuning and keypoint filtering during evaluation further improves robustness. Our method, evaluated on several datasets, achieves the best results in robot pose and joint angle estimation while being the least sensitive to occlusions and requiring the lowest execution time.
Atmospheric turbulence is a major source of image degradation in long-range imaging systems. Although numerous deep learning-based turbulence mitigation (TM) methods have been proposed, many are slow, memory-hungry, and do not generalize well. In the spatial domain, methods based on convolutional operators have a limited receptive field, so they cannot handle a large spatial dependency required by turbulence. In the temporal domain, methods relying on self-attention can, in theory, leverage the lucky effects of turbulence, but their quadratic complexity makes it difficult to scale to many frames. Traditional recurrent aggregation methods face parallelization challenges. In this paper, we present a new TM method based on two concepts: (1) A turbulence mitigation network based on the Selective State Space Model (MambaTM). MambaTM provides a global receptive field in each layer across spatial and temporal dimensions while maintaining linear computational complexity. (2) Learned Latent Phase Distortion (LPD). LPD guides the state space model. Unlike classical Zernike-based representations of phase distortion, the new LPD map uniquely captures the actual effects of turbulence, significantly improving the model's capability to estimate degradation by reducing the ill-posedness. Our proposed method exceeds current state-of-the-art networks on various synthetic and real-world TM benchmarks with significantly faster inference speed. The code will be publicly available.
UCM-VeID V2: A Richer Dataset and A Pre-training Method for UAV Cross-Modality Vehicle Re-Identification
PDF ↗Cross-Modality Re-Identification (VI-ReID) aims to achieve around-the-clock target matching, benefiting from the strengths of both RGB and infrared (IR) modalities. However, the field is hindered by limited datasets, particularly for vehicle VI-ReID, and by challenges such as modality bias training (MBT), stemming from biased pre-training on ImageNet. To tackle the above issues, this paper introduces an UCM-VeID V2 dataset benchmark for vehicle VI-ReID, and proposes a new self-supervised pre-training method, Cross-Modality Patch-Mixed Self-supervised Learning (PMSL). UCM-VeID V2 dataset features a significant increase in data volume, along with enhancements in multiple aspects. PMSL addresses MBT by learning modality-invariant features through Patch-Mixed Image Reconstruction (PMIR) and Modality Discrimination Adversarial Learning (MDAL), and enhances discriminability with Modality-Augmented Contrasting Cluster (MACC). Comprehensive experiments are carried out to validate the effectiveness of the proposed method.
Text serves as the key control signal in video generation due to its narrative nature. To render text descriptions into video clips, current video diffusion models borrow features from text encoders yet struggle with limited text comprehension. The recent success of large language models (LLMs) showcases the power of decoder-only transformers, which offers three clear benefits for text-to-video (T2V) generation, namely, precise text understanding resulting from the superior scalability, imagination beyond the input text enabled by next token prediction, and flexibility to prioritize user interests through instruction tuning. Nevertheless, the feature distribution gap emerging from the two different text modeling paradigms hinders the direct use of LLMs in established T2V models. This work addresses this challenge with Mimir, an end-to-end training framework featuring a carefully tailored token fuser to harmonize the outputs from text encoders and LLMs. Such a design allows the T2V model to fully leverage learned video priors while capitalizing on the text-related capability of LLMs. Extensive quantitative and qualitative results demonstrate the effectiveness of our approach in generating high-quality videos with excellent text comprehension, especially when processing short captions and managing shifting motions. The code and models will be made publicly available.
Contrastive loss is a powerful approach for representation learning, where larger batch sizes enhance performance by providing more negative samples to better distinguish between similar and dissimilar data. However, the full instantiation of the similarity matrix demands substantial GPU memory, making large batch training highly resource-intensive. To address this, we propose a tile-based computation strategy that partitions the contrastive loss calculation into small blocks, avoiding full materialization of the similarity matrix. Additionally, we introduce a multi-level tiling implementation to leverage the hierarchical structure of distributed systems, using ring-based communication at the GPU level to optimize synchronization and fused kernels at the CUDA core level to reduce I/O overhead. Experimental results show that the proposed method significantly reduces GPU memory usage in contrastive loss. For instance, it enables contrastive training of a CLIP-ViT-L/14 model with a batch size of 4M using only 8 A800 80GB GPUs, without sacrificing accuracy. Compared to state-of-the-art memory-efficient solutions, it achieves a two-order-of-magnitude reduction in memory while maintaining comparable speed. The code will be made publicly available.
Blind Super-Resolution(blind SR) aims to enhance the model's generalization ability with unknown degradation, yet it still encounters severe overfitting issues. Some previous methods inspired by dropout, which enhances generalization by regularizing features, have shown promising results in blind SR. Nevertheless, these methods focus solely on regularizing features before the final layer and overlook the need for generalization in features at intermediate layers. Without explicit regularization of features at intermediate layers, the blind SR network struggles to obtain well-generalized feature representations. However, the key challenge is that directly applying dropout to intermediate layers leads to a significant performance drop, which we attribute to the inconsistency in training-testing and across layers it introduced. Therefore, we propose Adaptive Dropout, a new regularization method for blind SR models, which mitigates the inconsistency and facilitates application across intermediate layers of networks. Specifically, for training-testing inconsistency, we re-design the form of dropout and integrate the features before and after dropout adaptively. For inconsistency in generalization requirements across different layers, we innovatively design an adaptive training strategy to strengthen feature propagation by layer-wise annealing. Experimental results show that our method outperforms all past regularization methods on both synthetic and real-world benchmark datasets, also highly effective in other image restoration tasks.
Prior research on out-of-distribution detection (OoDD) has primarily focused on single-modality models. Recently, with the advent of large-scale pretrained vision-language models such as CLIP, OoDD methods utilizing such multi-modal representations through zero-shot and prompt learning strategies have emerged. However, these methods typically involve either freezing the pretrained weights or only partially tuning them, which can be suboptimal for downstream datasets. In this paper, we highlight that multi-modal fine-tuning (MMFT) can achieve notable OoDD performance. Despite some recent works demonstrating the impact of fine-tuning methods for OoDD, there remains significant potential for performance improvement. We investigate the limitation of naive fine-tuning methods, examining why they fail to fully leverage the pretrained knowledge. Our empirical analysis suggests that this issue could stem from the modality gap within in-distribution (ID) embeddings. To address this, we propose a training objective that enhances cross-modal alignment by regularizing the distances between image and text embeddings of ID data. This adjustment helps in better utilizing pretrained textual information by aligning similar semantics from different modalities (i.e., text and image) more closely in the hyperspherical representation space. We theoretically demonstrate that the proposed regularization corresponds to the maximum likelihood estimation of an energy-based model on a hypersphere. Utilizing ImageNet-1k OoD benchmark datasets, we show that our method, combined with post-hoc OoDD approaches leveraging pretrained knowledge (e.g., NegLabel), significantly outperforms existing methods, achieving state-of-the-art OoDD performance and leading ID accuracy.
Recent works have shown that visual pretraining on egocentric datasets using masked autoencoders (MAE) can improve generalization for downstream robotics tasks. However, these approaches pretrain only on 2D images, while many robotics applications require 3D scene understanding. In this work, we propose 3D-MVP, a novel approach for 3D Multi-View Pretraining using masked autoencoders. We leverage Robotic View Transformer (RVT), which uses a multi-view transformer to understand the 3D scene and predict gripper pose actions. We split RVT's multi-view transformer into visual encoder and action decoder, and pretrain its visual encoder using masked autoencoding on large-scale 3D datasets such as Objaverse. We evaluate 3D-MVP on a suite of virtual robot manipulation tasks and demonstrate improved performance over baselines. Our results suggest that 3D-aware pretraining is a promising approach to improve generalization of vision-based robotic manipulation policies.
World models that forecast environmental changes from actions are vital for autonomous driving models with strong generalization. The prevailing driving world model mainly build on pixel-level video prediction model. Although these models can produce high-fidelity video sequences with advanced diffusion-based generator, they are constrained by their predictive duration and overall generalization capabilities. In this paper, we explore to solve this problem by combining pixel-level generation loss with MAE-style feature-level context learning. In particular, we instantiate this target with three key design: (1) A more scalable Diffusion Transformer (DiT) structure trained with extra mask construction task. (2) we devise diffusion-related mask tokens to deal with the fuzzy relations between mask reconstruction and generative diffusion process. (3) we extend mask construction task to spatial-temporal domain by utilizing row-wise mask for shifted self-attention rather than masked self-attention in MAE. Then, we adopt a row-wise cross-view module to align with this mask design. Based on above improvement, we propose MaskGWM: a Generalizable driving World Model embodied with Video Mask reconstruction. Our model contains two variants: MaskGWM-long, focusing on long-horizon prediction, and MaskGWM-mview, dedicated to multi-view generation.Comprehensive experiments on standard benchmarks validate the effectiveness of the proposed method, which contain normal validation of Nuscene dataset, long-horizon rollout of OpenDV-2K dataset and zero-shot validation of Waymo dataset. Quantitative metrics on these datasets show our method notably improving state-of-the-art driving world model.
VISCO: Benchmarking Fine-Grained Critique and Correction Towards Self-Improvement in Visual Reasoning
PDF ↗The ability of large vision-language models (LVLMs) to critique and correct their reasoning is an essential building block towards their self-improvement. However, a systematic analysis of such capabilities in LVLMs is still lacking. We propose VISCO, the first benchmark to extensively analyze the fine-grained critique and correction capabilities of LVLMs. Compared to existing work that uses a single scalar value to critique the entire reasoning [4], VISCO features **dense** and **fine-grained** critique, requiring LVLMs to evaluate the correctness of each step in the chain-of-thought and provide natural language explanations to support their judgments. Extensive evaluation of 24 LVLMs demonstrates that human-written critiques significantly enhance the performance after correction, showcasing the potential of the self-improvement strategy. However, the model-generated critiques are less helpful and sometimes detrimental to the performance, suggesting that critique is the crucial bottleneck. We identified three common patterns in critique failures: failure to critique visual perception, reluctance to "say no", and exaggerated assumption of error propagation. To address these issues, we propose an effective LookBack strategy that revisits the image to verify each piece of information in the initial reasoning. \ourscritic significantly improves critique and correction performance by up to 13.5%.
Geodesic distance serves as a reliable means of measuring distance in nonlinear spaces, and such nonlinear manifolds are prevalent in the current multimodal learning. In these scenarios, some samples may exhibit high similarity, yet they convey different semantics, making traditional distance metrics inadequate for distinguishing between positive and negative samples. This paper introduces geodesic distance as a novel distance metric in multi-modal learning for the first time, to mine correlations between samples, aiming to address the limitations of common distance metric. Our approach incorporates a comprehensive series of strategies to adapt geodesic distance for the current multimodal learning. Specifically, we construct a graph structure to represent the adjacency relationships among samples by thresholding distances between them and then apply the shortest-path algorithm to obtain geodesic distance within this graph. To facilitate efficient computation, we further propose a hierarchical graph structure through clustering and combined with incremental update strategies for dynamic status updates. Extensive experiments across various downstream tasks validate the effectiveness of our proposed method, demonstrating its capability to capture complex relationships between samples and improve the performance of multimodal learning models.
We introduce UniGraspTransformer, a universal Transformer-based network for dexterous robotic grasping that simplifies training while enhancing scalability and performance. Unlike prior methods such as UniDexGrasp++, which require complex, multi-step training pipelines, UniGraspTransformer follows a streamlined process: first, dedicated policy networks are trained for individual objects using reinforcement learning to generate successful grasp trajectories; then, these trajectories are distilled into a single, universal network. Our approach enables UniGraspTransformer to scale effectively, incorporating up to 12 self-attention blocks for handling thousands of objects with diverse poses. Additionally, it generalizes well to both idealized and real-world inputs, evaluated in state-based and vision-based settings. Notably, UniGraspTransformer generates a broader range of grasping poses for objects in various shapes and orientations, resulting in more diverse grasp strategies. Experimental results demonstrate significant improvements over state-of-the-art, UniDexGrasp++, across various object categories, achieving success rate gains of 3.5%, 7.7%, and 10.1% on seen objects, unseen objects within seen categories, and completely unseen objects, respectively, in the vision-based setting. Project page: https://dexhand.github.io/UniGraspTransformer/
Reason-before-Retrieve: One-Stage Reflective Chain-of-Thoughts for Training-Free Zero-Shot Composed Image Retrieval
PDF ↗Composed Image Retrieval (CIR) aims to retrieve target images that closely resemble a reference image while integrating user-specified textual modifications, thereby capturing user intent more accurately. Existing training-free zero-shot CIR (ZS-CIR) methods often employ a two-stage process: they first generate a caption for the reference image and then use Large Language Models for reasoning a target description. However, these methods suffer from missing critical visual details and limited reasoning capabilities, leading to suboptimal retrieval performance. To address these challenges, we propose a novel, training-free one-stage method, One-Stage Reflective Chain-of-Thought Reasoning (OSrCIR) for ZS-CIR, which employs Multimodal Large Language Models to retain essential visual information in a single-stage reasoning process, eliminating the information loss in two-stage methods. Our Reflective Chain-of-Thought framework further improves interpretative accuracy by aligning manipulation intent with contextual cues from reference images. OSrCIR achieves performance gains of 1.80% to 6.44% over existing training-free methods across multiple tasks, setting new state-of-the-art results in ZS-CIR and enhancing its utility in vision-language applications. Our code is available at https://github.com/microsoft/ACV/tree/main/OSrCIR.
Many real-world applications of flow-based generative models desire a diverse set of samples that cover multiple modes of the target distribution. However, the predominant approach for obtaining diverse sets is not sample-efficient, as it involves independently obtaining many samples from the source distribution and mapping them through the flow until the desired mode coverage is achieved. As an alternative to repeated sampling, we introduce DiverseFlow: a training-free approach to improve the diversity of flow models. Our key idea is to employ a determinantal point process to induce a coupling between the samples that drives diversity under a fixed sampling budget. In essence, DiverseFlow allows exploration of more variations in a learned flow model with fewer samples. We demonstrate the efficacy of our method for tasks where sample-efficient diversity is desirable, such as text-guided image generation with polysemous words, inverse problems like large-hole inpainting, and class-conditional image synthesis.
Touch contact and pressure are essential for understanding how humans interact with objects and offer insights that benefit applications in mixed reality and robotics. Estimating these interactions from an egocentric camera perspective is challenging, largely due to the lack of comprehensive datasets that provide both hand poses and pressure annotations. In this paper, we present EgoPressure, an egocentric dataset that is annotated with high-resolution pressure intensities at contact points and precise hand pose meshes, obtained via our multi-view, sequence-based optimization method. We introduce baseline models for estimating applied pressure on external surfaces from RGB images, both with and without hand pose information, as well as a joint model for predicting hand pose and the pressure distribution across the hand mesh.Our experiments show that pressure and hand pose complement each other in understanding hand-object interactions.
MoFlow: One-Step Flow Matching for Human Trajectory Forecasting via Implicit Maximum Likelihood Estimation based Distillation
PDF ↗In this paper, we address the problem of human trajectory forecasting, which aims to predict the inherently multi-modal future movements of humans based on their past trajectories and other contextual cues. We propose a novel motion prediction conditional flow matching model, termed MoFlow, to predict K-shot future trajectories for all agents in a given scene. We design a novel flow matching loss function that not only ensures at least one of the K sets of future trajectories is accurate but also encourages all K sets of future trajectories to be diverse and plausible. Furthermore, by leveraging the implicit maximum likelihood estimation (IMLE), we propose a novel distillation method for flow models that only requires samples from the teacher model. Extensive experiments on the real-world datasets, including SportVU NBA games, ETH-UCY, and SDD, demonstrate that both our teacher flow model and the IMLE-distilled student model achieve state-of-the-art performance. These models can generate diverse trajectories that are physically and socially plausible. Moreover, our one-step student model is 100 times faster than the teacher flow model during sampling. The code, model, and data are available at our project page: https://moflow-imle.github.io/.
Non-exemplar Class Incremental Learning (NECIL) enables models to continuously acquire new classes without retraining from scratch and storing old task exemplars, addressing privacy and storage issues. However, the absence of data from earlier tasks exacerbates the challenge of catastrophic forgetting in NECIL. In this paper, we propose a novel framework called Dynamic Integration of task-specific Adapters (DIA), which comprises two key components: Task-Specific Adapter Integration (TSAI) and Patch-Level Model Alignment. TSAI boosts compositionality through a patch-level adapter integration strategy, aggregating richer task-specific information while maintaining low computation costs. Patch-Level Model Alignment maintains feature consistency and accurate decision boundaries via two specialized mechanisms: Patch-Level Distillation Loss (PDL) and Patch-Level Feature Reconstruction (PFR). Specifically, on the one hand, the PDL preserves feature-level consistency between successive models by implementing a distillation loss based on the contributions of patch tokens to new class learning. On the other hand, the PFR promotes classifier alignment by reconstructing old class features from previous tasks that adapt to new task knowledge, thereby preserving well-calibrated decision boundaries. Comprehensive experiments validate the effectiveness of our DIA, revealing significant improvements on NECIL benchmark datasets while maintaining an optimal balance between computational complexity and accuracy.
Simulator HC: Regression-based Online Simulation of Starting Problem-Solution Pairs for Homotopy Continuation in Geometric Vision
PDF ↗While automatically generated polynomial elimination templates have sparked great progress in the field of 3D computer vision, there remain many problems for which the degree of the constraints or the number of unknowns leads to intractability. In recent years, homotopy continuation has been introduced as a plausible alternative. However, the method currently depends on expensive parallel tracking of all possible solutions in the complex domain, or a classification network for starting problem-solution pairs trained over a limited set of real-world examples. Our innovation lies in a novel approach to finding solution-problem pairs, where we only need to predict a rough initial solution, with the corresponding problem generated by an online simulator. Subsequently, homotopy continuation is applied to track that single solution back to the original problem. We apply this elegant combination to generalized camera resectioning, and also introduce a new solution to the challenging generalized relative pose and scale problem. As demonstrated, the proposed method successfully compensates the raw error committed by the regressor alone, and leads to state-of-the-art efficiency and success rates.
We present a snapshot imaging technique for recovering 3D surrounding views of miniature scenes. Due to their intricacy, miniature scenes with objects sized in millimeters are difficult to reconstruct, yet miniatures are common in life and their 3D digitalization is desirable. We design a catadioptric imaging system with a single camera and eight pairs of planar mirrors for snapshot 3D reconstruction from a dollhouse perspective. We place paired mirrors on nested pyramid surfaces for capturing surrounding multi-view images in a single shot. Our mirror design is customizable based on the size of the scene for optimized view coverage. We use the 3D Gaussian Splatting (3DGS) representation for scene reconstruction and novel view synthesis. We overcome the challenge posed by our sparse view input by integrating visual hull-derived depth constraint. Our method demonstrates state-of-the-art performance on a variety of synthetic and real miniature scenes.