Tsinghua University
Abstract:Vision foundation models are increasingly employed in autonomous driving systems due to their advanced capabilities. However, these models are susceptible to adversarial attacks, posing significant risks to the reliability and safety of autonomous vehicles. Adversaries can exploit these vulnerabilities to manipulate the vehicle's perception of its surroundings, leading to erroneous decisions and potentially catastrophic consequences. To address this challenge, we propose a novel Precision-Guided Adversarial Attack (PG-Attack) framework that combines two techniques: Precision Mask Perturbation Attack (PMP-Attack) and Deceptive Text Patch Attack (DTP-Attack). PMP-Attack precisely targets the attack region to minimize the overall perturbation while maximizing its impact on the target object's representation in the model's feature space. DTP-Attack introduces deceptive text patches that disrupt the model's understanding of the scene, further enhancing the attack's effectiveness. Our experiments demonstrate that PG-Attack successfully deceives a variety of advanced multi-modal large models, including GPT-4V, Qwen-VL, and imp-V1. Additionally, we won First-Place in the CVPR 2024 Workshop Challenge: Black-box Adversarial Attacks on Vision Foundation Models and codes are available at https://github.com/fuhaha824/PG-Attack.
Abstract:Multimodal Large Models (MLMs) are becoming a significant research focus, combining powerful large language models with multimodal learning to perform complex tasks across different data modalities. This review explores the latest developments and challenges in MLMs, emphasizing their potential in achieving artificial general intelligence and as a pathway to world models. We provide an overview of key techniques such as Multimodal Chain of Thought (M-COT), Multimodal Instruction Tuning (M-IT), and Multimodal In-Context Learning (M-ICL). Additionally, we discuss both the fundamental and specific technologies of multimodal models, highlighting their applications, input/output modalities, and design characteristics. Despite significant advancements, the development of a unified multimodal model remains elusive. We discuss the integration of 3D generation and embodied intelligence to enhance world simulation capabilities and propose incorporating external rule systems for improved reasoning and decision-making. Finally, we outline future research directions to address these challenges and advance the field.
Abstract:The contemporary state-of-the-art of Dynamic Facial Expression Recognition (DFER) technology facilitates remarkable progress by deriving emotional mappings of facial expressions from video content, underpinned by training on voluminous datasets. Yet, the DFER datasets encompass a substantial volume of noise data. Noise arises from low-quality captures that defy logical labeling, and instances that suffer from mislabeling due to annotation bias, engendering two principal types of uncertainty: the uncertainty regarding data usability and the uncertainty concerning label reliability. Addressing the two types of uncertainty, we have meticulously crafted a two-stage framework aiming at \textbf{S}eeking \textbf{C}ertain data \textbf{I}n extensive \textbf{U}ncertain data (SCIU). This initiative aims to purge the DFER datasets of these uncertainties, thereby ensuring that only clean, verified data is employed in training processes. To mitigate the issue of low-quality samples, we introduce the Coarse-Grained Pruning (CGP) stage, which assesses sample weights and prunes those deemed unusable due to their low weight. For samples with incorrect annotations, the Fine-Grained Correction (FGC) stage evaluates prediction stability to rectify mislabeled data. Moreover, SCIU is conceived as a universally compatible, plug-and-play framework, tailored to integrate seamlessly with prevailing DFER methodologies. Rigorous experiments across prevalent DFER datasets and against numerous benchmark methods substantiates SCIU's capacity to markedly elevate performance metrics.
Abstract:The problem of blind image super-resolution aims to recover high-resolution (HR) images from low-resolution (LR) images with unknown degradation modes. Most existing methods model the image degradation process using blur kernels. However, this explicit modeling approach struggles to cover the complex and varied degradation processes encountered in the real world, such as high-order combinations of JPEG compression, blur, and noise. Implicit modeling for the degradation process can effectively overcome this issue, but a key challenge of implicit modeling is the lack of accurate ground truth labels for the degradation process to conduct supervised training. To overcome this limitations inherent in implicit modeling, we propose an \textbf{U}ncertainty-based degradation representation for blind \textbf{S}uper-\textbf{R}esolution framework (\textbf{USR}). By suppressing the uncertainty of local degradation representations in images, USR facilitated self-supervised learning of degradation representations. The USR consists of two components: Adaptive Uncertainty-Aware Degradation Extraction (AUDE) and a feature extraction network composed of Variable Depth Dynamic Convolution (VDDC) blocks. To extract Uncertainty-based Degradation Representation from LR images, the AUDE utilizes the Self-supervised Uncertainty Contrast module with Uncertainty Suppression Loss to suppress the inherent model uncertainty of the Degradation Extractor. Furthermore, VDDC block integrates degradation information through dynamic convolution. Rhe VDDC also employs an Adaptive Intensity Scaling operation that adaptively adjusts the degradation representation according to the network hierarchy, thereby facilitating the effective integration of degradation information. Quantitative and qualitative experiments affirm the superiority of our approach.
Abstract:Dynamic Facial Expression Recognition (DFER) is crucial for affective computing but often overlooks the impact of scene context. We have identified a significant issue in current DFER tasks: human annotators typically integrate emotions from various angles, including environmental cues and body language, whereas existing DFER methods tend to consider the scene as noise that needs to be filtered out, focusing solely on facial information. We refer to this as the Rigid Cognitive Problem. The Rigid Cognitive Problem can lead to discrepancies between the cognition of annotators and models in some samples. To align more closely with the human cognitive paradigm of emotions, we propose an Overall Understanding of the Scene DFER method (OUS). OUS effectively integrates scene and facial features, combining scene-specific emotional knowledge for DFER. Extensive experiments on the two largest datasets in the DFER field, DFEW and FERV39k, demonstrate that OUS significantly outperforms existing methods. By analyzing the Rigid Cognitive Problem, OUS successfully understands the complex relationship between scene context and emotional expression, closely aligning with human emotional understanding in real-world scenarios.
Abstract:Video object segmentation (VOS) aims to distinguish and track target objects in a video. Despite the excellent performance achieved by off-the-shell VOS models, existing VOS benchmarks mainly focus on short-term videos lasting about 5 seconds, where objects remain visible most of the time. However, these benchmarks poorly represent practical applications, and the absence of long-term datasets restricts further investigation of VOS in realistic scenarios. Thus, we propose a novel benchmark named LVOS, comprising 720 videos with 296,401 frames and 407,945 high-quality annotations. Videos in LVOS last 1.14 minutes on average, approximately 5 times longer than videos in existing datasets. Each video includes various attributes, especially challenges deriving from the wild, such as long-term reappearing and cross-temporal similar objects. Compared to previous benchmarks, our LVOS better reflects VOS models' performance in real scenarios. Based on LVOS, we evaluate 20 existing VOS models under 4 different settings and conduct a comprehensive analysis. On LVOS, these models suffer a large performance drop, highlighting the challenge of achieving precise tracking and segmentation in real-world scenarios. Attribute-based analysis indicates that key factor to accuracy decline is the increased video length, emphasizing LVOS's crucial role. We hope our LVOS can advance development of VOS in real scenes. Data and code are available at https://lingyihongfd.github.io/lvos.github.io/.
Abstract:Data-Free Knowledge Distillation (DFKD) is a promising task to train high-performance small models to enhance actual deployment without relying on the original training data. Existing methods commonly avoid relying on private data by utilizing synthetic or sampled data. However, a long-overlooked issue is that the severe distribution shifts between their substitution and original data, which manifests as huge differences in the quality of images and class proportions. The harmful shifts are essentially the confounder that significantly causes performance bottlenecks. To tackle the issue, this paper proposes a novel perspective with causal inference to disentangle the student models from the impact of such shifts. By designing a customized causal graph, we first reveal the causalities among the variables in the DFKD task. Subsequently, we propose a Knowledge Distillation Causal Intervention (KDCI) framework based on the backdoor adjustment to de-confound the confounder. KDCI can be flexibly combined with most existing state-of-the-art baselines. Experiments in combination with six representative DFKD methods demonstrate the effectiveness of our KDCI, which can obviously help existing methods under almost all settings, \textit{e.g.}, improving the baseline by up to 15.54\% accuracy on the CIFAR-100 dataset.
Abstract:In software evolution, resolving the emergent issues within GitHub repositories is a complex challenge that involves not only the incorporation of new code but also the maintenance of existing functionalities. Large Language Models (LLMs) have shown promise in code generation and understanding but face difficulties in code change, particularly at the repository level. To overcome these challenges, we empirically study the reason why LLMs mostly fail to resolve GitHub issues and analyze some impact factors. Motivated by the empirical findings, we propose a novel LLM-based Multi-Agent framework for GitHub Issue reSolution, MAGIS, consisting of four kinds of agents customized for the software evolution: Manager, Repository Custodian, Developer, and Quality Assurance Engineer agents. This framework leverages the collaboration of various agents in the planning and coding process to unlock the potential of LLMs to resolve GitHub issues. In experiments, we employ the SWE-bench benchmark to compare MAGIS with popular LLMs, including GPT-3.5, GPT-4, and Claude-2. MAGIS can resolve 13.94% GitHub issues, which significantly outperforms the baselines. Specifically, MAGIS achieves an eight-fold increase in resolved ratio over the direct application of GPT-4, the based LLM of our method. We also analyze the factors for improving GitHub issue resolution rates, such as line location, task allocation, etc.
Abstract:Despite the substantial advancements in Vision-Language Pre-training (VLP) models, their susceptibility to adversarial attacks poses a significant challenge. Existing work rarely studies the transferability of attacks on VLP models, resulting in a substantial performance gap from white-box attacks. We observe that prior work overlooks the interaction mechanisms between modalities, which plays a crucial role in understanding the intricacies of VLP models. In response, we propose a novel attack, called Collaborative Multimodal Interaction Attack (CMI-Attack), leveraging modality interaction through embedding guidance and interaction enhancement. Specifically, attacking text at the embedding level while preserving semantics, as well as utilizing interaction image gradients to enhance constraints on perturbations of texts and images. Significantly, in the image-text retrieval task on Flickr30K dataset, CMI-Attack raises the transfer success rates from ALBEF to TCL, $\text{CLIP}_{\text{ViT}}$ and $\text{CLIP}_{\text{CNN}}$ by 8.11%-16.75% over state-of-the-art methods. Moreover, CMI-Attack also demonstrates superior performance in cross-task generalization scenarios. Our work addresses the underexplored realm of transfer attacks on VLP models, shedding light on the importance of modality interaction for enhanced adversarial robustness.
Abstract:Visual object tracking aims to localize the target object of each frame based on its initial appearance in the first frame. Depending on the input modility, tracking tasks can be divided into RGB tracking and RGB+X (e.g. RGB+N, and RGB+D) tracking. Despite the different input modalities, the core aspect of tracking is the temporal matching. Based on this common ground, we present a general framework to unify various tracking tasks, termed as OneTracker. OneTracker first performs a large-scale pre-training on a RGB tracker called Foundation Tracker. This pretraining phase equips the Foundation Tracker with a stable ability to estimate the location of the target object. Then we regard other modality information as prompt and build Prompt Tracker upon Foundation Tracker. Through freezing the Foundation Tracker and only adjusting some additional trainable parameters, Prompt Tracker inhibits the strong localization ability from Foundation Tracker and achieves parameter-efficient finetuning on downstream RGB+X tracking tasks. To evaluate the effectiveness of our general framework OneTracker, which is consisted of Foundation Tracker and Prompt Tracker, we conduct extensive experiments on 6 popular tracking tasks across 11 benchmarks and our OneTracker outperforms other models and achieves state-of-the-art performance.