Dongfang Liu

Cheng Han, Qifan Wang, Yiming Cui et al.

As the size of transformer-based models continues to grow, fine-tuning these large-scale pretrained vision models for new tasks has become increasingly parameter-intensive. Parameter-efficient learning has been developed to reduce the number of tunable parameters during fine-tuning. Although these methods show promising results, there is still a significant performance gap compared to full fine-tuning. To address this challenge, we propose an Effective and Efficient Visual Prompt Tuning (E^2VPT) approach for large-scale transformer-based model adaptation. Specifically, we introduce a set of learnable key-value prompts and visual prompts into self-attention and input layers, respectively, to improve the effectiveness of model fine-tuning. Moreover, we design a prompt pruning procedure to systematically prune low importance prompts while preserving model performance, which largely enhances the model's efficiency. Empirical results demonstrate that our approach outperforms several state-of-the-art baselines on two benchmarks, with considerably low parameter usage (e.g., 0.32% of model parameters on VTAB-1k). Our code is available at https://github.com/ChengHan111/E2VPT.

Liqi Yan, Qifan Wang, Yiming Cui et al.

Video captioning is a challenging task as it needs to accurately transform visual understanding into natural language description. To date, state-of-the-art methods inadequately model global-local representation across video frames for caption generation, leaving plenty of room for improvement. In this work, we approach the video captioning task from a new perspective and propose a GL-RG framework for video captioning, namely a \textbf{G}lobal-\textbf{L}ocal \textbf{R}epresentation \textbf{G}ranularity. Our GL-RG demonstrates three advantages over the prior efforts: 1) we explicitly exploit extensive visual representations from different video ranges to improve linguistic expression; 2) we devise a novel global-local encoder to produce rich semantic vocabulary to obtain a descriptive granularity of video contents across frames; 3) we develop an incremental training strategy which organizes model learning in an incremental fashion to incur an optimal captioning behavior. Experimental results on the challenging MSR-VTT and MSVD datasets show that our DL-RG outperforms recent state-of-the-art methods by a significant margin. Code is available at \url{https://github.com/ylqi/GL-RG}.

Zhiyuan Cheng, Hongjun Choi, James Liang et al.

Multi-sensor fusion (MSF) is widely used in autonomous vehicles (AVs) for perception, particularly for 3D object detection with camera and LiDAR sensors. The purpose of fusion is to capitalize on the advantages of each modality while minimizing its weaknesses. Advanced deep neural network (DNN)-based fusion techniques have demonstrated the exceptional and industry-leading performance. Due to the redundant information in multiple modalities, MSF is also recognized as a general defence strategy against adversarial attacks. In this paper, we attack fusion models from the camera modality that is considered to be of lesser importance in fusion but is more affordable for attackers. We argue that the weakest link of fusion models depends on their most vulnerable modality, and propose an attack framework that targets advanced camera-LiDAR fusion-based 3D object detection models through camera-only adversarial attacks. Our approach employs a two-stage optimization-based strategy that first thoroughly evaluates vulnerable image areas under adversarial attacks, and then applies dedicated attack strategies for different fusion models to generate deployable patches. The evaluations with six advanced camera-LiDAR fusion models and one camera-only model indicate that our attacks successfully compromise all of them. Our approach can either decrease the mean average precision (mAP) of detection performance from 0.824 to 0.353, or degrade the detection score of a target object from 0.728 to 0.156, demonstrating the efficacy of our proposed attack framework. Code is available.

Wenguan Wang, Cheng Han, Tianfei Zhou et al.

We devise deep nearest centroids (DNC), a conceptually elegant yet surprisingly effective network for large-scale visual recognition, by revisiting Nearest Centroids, one of the most classic and simple classifiers. Current deep models learn the classifier in a fully parametric manner, ignoring the latent data structure and lacking simplicity and explainability. DNC instead conducts nonparametric, case-based reasoning; it utilizes sub-centroids of training samples to describe class distributions and clearly explains the classification as the proximity of test data and the class sub-centroids in the feature space. Due to the distance-based nature, the network output dimensionality is flexible, and all the learnable parameters are only for data embedding. That means all the knowledge learnt for ImageNet classification can be completely transferred for pixel recognition learning, under the "pre-training and fine-tuning" paradigm. Apart from its nested simplicity and intuitive decision-making mechanism, DNC can even possess ad-hoc explainability when the sub-centroids are selected as actual training images that humans can view and inspect. Compared with parametric counterparts, DNC performs better on image classification (CIFAR-10, ImageNet) and greatly boots pixel recognition (ADE20K, Cityscapes), with improved transparency and fewer learnable parameters, using various network architectures (ResNet, Swin) and segmentation models (FCN, DeepLabV3, Swin). We feel this work brings fundamental insights into related fields.

Guangyan Sun, Mingyu Jin, Zhenting Wang et al.

Achieving human-level intelligence requires refining cognitive distinctions between System 1 and System 2 thinking. While contemporary AI, driven by large language models, demonstrates human-like traits, it falls short of genuine cognition. Transitioning from structured benchmarks to real-world scenarios presents challenges for visual agents, often leading to inaccurate and overly confident responses. To address the challenge, we introduce FaST, which incorporates the Fast and Slow Thinking mechanism into visual agents. FaST employs a switch adapter to dynamically select between System 1/2 modes, tailoring the problem-solving approach to different task complexity. It tackles uncertain and unseen objects by adjusting model confidence and integrating new contextual data. With this novel design, we advocate a flexible system, hierarchical reasoning capabilities, and a transparent decision-making pipeline, all of which contribute to its ability to emulate human-like cognitive processes in visual intelligence. Empirical results demonstrate that FaST outperforms various well-known baselines, achieving 80.8% accuracy over VQA^{v2} for visual question answering and 48.7% GIoU score over ReasonSeg for reasoning segmentation, demonstrate FaST's superior performance. Extensive testing validates the efficacy and robustness of FaST's core components, showcasing its potential to advance the development of cognitive visual agents in AI systems. The code is available at ttps://github.com/GuangyanS/Sys2-LLaVA.

James C. Liang, Yiming Cui, Qifan Wang et al.

This paper presents CLUSTERFORMER, a universal vision model that is based on the CLUSTERing paradigm with TransFORMER. It comprises two novel designs: 1. recurrent cross-attention clustering, which reformulates the cross-attention mechanism in Transformer and enables recursive updates of cluster centers to facilitate strong representation learning; and 2. feature dispatching, which uses the updated cluster centers to redistribute image features through similarity-based metrics, resulting in a transparent pipeline. This elegant design streamlines an explainable and transferable workflow, capable of tackling heterogeneous vision tasks (i.e., image classification, object detection, and image segmentation) with varying levels of clustering granularity (i.e., image-, box-, and pixel-level). Empirical results demonstrate that CLUSTERFORMER outperforms various well-known specialized architectures, achieving 83.41% top-1 acc. over ImageNet-1K for image classification, 54.2% and 47.0% mAP over MS COCO for object detection and instance segmentation, 52.4% mIoU over ADE20K for semantic segmentation, and 55.8% PQ over COCO Panoptic for panoptic segmentation. For its efficacy, we hope our work can catalyze a paradigm shift in universal models in computer vision.

Zhiyuan Cheng, James Liang, Hongjun Choi et al.

Deep learning has substantially boosted the performance of Monocular Depth Estimation (MDE), a critical component in fully vision-based autonomous driving (AD) systems (e.g., Tesla and Toyota). In this work, we develop an attack against learning-based MDE. In particular, we use an optimization-based method to systematically generate stealthy physical-object-oriented adversarial patches to attack depth estimation. We balance the stealth and effectiveness of our attack with object-oriented adversarial design, sensitive region localization, and natural style camouflage. Using real-world driving scenarios, we evaluate our attack on concurrent MDE models and a representative downstream task for AD (i.e., 3D object detection). Experimental results show that our method can generate stealthy, effective, and robust adversarial patches for different target objects and models and achieves more than 6 meters mean depth estimation error and 93% attack success rate (ASR) in object detection with a patch of 1/9 of the vehicle's rear area. Field tests on three different driving routes with a real vehicle indicate that we cause over 6 meters mean depth estimation error and reduce the object detection rate from 90.70% to 5.16% in continuous video frames.

Yawen Lu, Qifan Wang, Siqi Ma et al.

Optical flow is an indispensable building block for various important computer vision tasks, including motion estimation, object tracking, and disparity measurement. In this work, we propose TransFlow, a pure transformer architecture for optical flow estimation. Compared to dominant CNN-based methods, TransFlow demonstrates three advantages. First, it provides more accurate correlation and trustworthy matching in flow estimation by utilizing spatial self-attention and cross-attention mechanisms between adjacent frames to effectively capture global dependencies; Second, it recovers more compromised information (e.g., occlusion and motion blur) in flow estimation through long-range temporal association in dynamic scenes; Third, it enables a concise self-learning paradigm and effectively eliminate the complex and laborious multi-stage pre-training procedures. We achieve the state-of-the-art results on the Sintel, KITTI-15, as well as several downstream tasks, including video object detection, interpolation and stabilization. For its efficacy, we hope TransFlow could serve as a flexible baseline for optical flow estimation.

Zhiyuan Cheng, James Liang, Guanhong Tao et al.

Monocular Depth Estimation (MDE) is a critical component in applications such as autonomous driving. There are various attacks against MDE networks. These attacks, especially the physical ones, pose a great threat to the security of such systems. Traditional adversarial training method requires ground-truth labels hence cannot be directly applied to self-supervised MDE that does not have ground-truth depth. Some self-supervised model hardening techniques (e.g., contrastive learning) ignore the domain knowledge of MDE and can hardly achieve optimal performance. In this work, we propose a novel adversarial training method for self-supervised MDE models based on view synthesis without using ground-truth depth. We improve adversarial robustness against physical-world attacks using L0-norm-bounded perturbation in training. We compare our method with supervised learning based and contrastive learning based methods that are tailored for MDE. Results on two representative MDE networks show that we achieve better robustness against various adversarial attacks with nearly no benign performance degradation.

Qifan Wang, Yi Fang, Anirudh Ravula et al.

Network embedding is an effective technique to learn the low-dimensional representations of nodes in networks. Real-world networks are usually with multiplex or having multi-view representations from different relations. Recently, there has been increasing interest in network embedding on multiplex data. However, most existing multiplex approaches assume that the data is complete in all views. But in real applications, it is often the case that each view suffers from the missing of some data and therefore results in partial multiplex data. In this paper, we present a novel Deep Partial Multiplex Network Embedding approach to deal with incomplete data. In particular, the network embeddings are learned by simultaneously minimizing the deep reconstruction loss with the autoencoder neural network, enforcing the data consistency across views via common latent subspace learning, and preserving the data topological structure within the same network through graph Laplacian. We further prove the orthogonal invariant property of the learned embeddings and connect our approach with the binary embedding techniques. Experiments on four multiplex benchmarks demonstrate the superior performance of the proposed approach over several state-of-the-art methods on node classification, link prediction and clustering tasks.

Zhiwen Cao, Dongfang Liu, Qifan Wang et al.

Facial pose estimation refers to the task of predicting face orientation from a single RGB image. It is an important research topic with a wide range of applications in computer vision. Label distribution learning (LDL) based methods have been recently proposed for facial pose estimation, which achieve promising results. However, there are two major issues in existing LDL methods. First, the expectations of label distributions are biased, leading to a biased pose estimation. Second, fixed distribution parameters are applied for all learning samples, severely limiting the model capability. In this paper, we propose an Anisotropic Spherical Gaussian (ASG)-based LDL approach for facial pose estimation. In particular, our approach adopts the spherical Gaussian distribution on a unit sphere which constantly generates unbiased expectation. Meanwhile, we introduce a new loss function that allows the network to learn the distribution parameter for each learning sample flexibly. Extensive experimental results show that our method sets new state-of-the-art records on AFLW2000 and BIWI datasets.

Yiming Cui, Cheng Han, Dongfang Liu

The advancement of computer vision has pushed visual analysis tasks from still images to the video domain. In recent years, video instance segmentation, which aims to track and segment multiple objects in video frames, has drawn much attention for its potential applications in various emerging areas such as autonomous driving, intelligent transportation, and smart retail. In this paper, we propose an effective framework for instance-level visual analysis on video frames, which can simultaneously conduct object detection, instance segmentation, and multi-object tracking. The core idea of our method is collaborative multi-task learning which is achieved by a novel structure, named associative connections among detection, segmentation, and tracking task heads in an end-to-end learnable CNN. These additional connections allow information propagation across multiple related tasks, so as to benefit these tasks simultaneously. We evaluate the proposed method extensively on KITTI MOTS and MOTS Challenge datasets and obtain quite encouraging results.

Cheng Han, Qifan Wang, Sohail A. Dianat et al.

Transformer-based architectures have become the de-facto standard models for diverse vision tasks owing to their superior performance. As the size of the models continues to scale up, model distillation becomes extremely important in various real applications, particularly on devices limited by computational resources. However, prevailing knowledge distillation methods exhibit diminished efficacy when confronted with a large capacity gap between the teacher and the student, e.g, 10x compression rate. In this paper, we present a novel approach named Automatic Multi-step Distillation (AMD) for large-scale vision model compression. In particular, our distillation process unfolds across multiple steps. Initially, the teacher undergoes distillation to form an intermediate teacher-assistant model, which is subsequently distilled further to the student. An efficient and effective optimization framework is introduced to automatically identify the optimal teacher-assistant that leads to the maximal student performance. We conduct extensive experiments on multiple image classification datasets, including CIFAR-10, CIFAR-100, and ImageNet. The findings consistently reveal that our approach outperforms several established baselines, paving a path for future knowledge distillation methods on large-scale vision models.

Liqi Yan, Qifan Wang, Junhan Zhao et al.

First-Person-View (FPV) holds immense potential for revolutionizing the trajectory of Unmanned Aerial Vehicles (UAVs), offering an exhilarating avenue for navigating complex building structures. Yet, traditional Neural Radiance Field (NeRF) methods face challenges such as sampling single points per iteration and requiring an extensive array of views for supervision. UAV videos exacerbate these issues with limited viewpoints and significant spatial scale variations, resulting in inadequate detail rendering across diverse scales. In response, we introduce FPV-NeRF, addressing these challenges through three key facets: (1) Temporal consistency. Leveraging spatio-temporal continuity ensures seamless coherence between frames; (2) Global structure. Incorporating various global features during point sampling preserves space integrity; (3) Local granularity. Employing a comprehensive framework and multi-resolution supervision for multi-scale scene feature representation tackles the intricacies of UAV video spatial scales. Additionally, due to the scarcity of publicly available FPV videos, we introduce an innovative view synthesis method using NeRF to generate FPV perspectives from UAV footage, enhancing spatial perception for drones. Our novel dataset spans diverse trajectories, from outdoor to indoor environments, in the UAV domain, differing significantly from traditional NeRF scenarios. Through extensive experiments encompassing both interior and exterior building structures, FPV-NeRF demonstrates a superior understanding of the UAV flying space, outperforming state-of-the-art methods in our curated UAV dataset. Explore our project page for further insights: https://fpv-nerf.github.io/.

Hongye Xu, Dongfang Liu, Cory Merkel et al.

Logic locking has been proposed to safeguard intellectual property (IP) during chip fabrication. Logic locking techniques protect hardware IP by making a subset of combinational modules in a design dependent on a secret key that is withheld from untrusted parties. If an incorrect secret key is used, a set of deterministic errors is produced in locked modules, restricting unauthorized use. A common target for logic locking is neural accelerators, especially as machine-learning-as-a-service becomes more prevalent. In this work, we explore how logic locking can be used to compromise the security of a neural accelerator it protects. Specifically, we show how the deterministic errors caused by incorrect keys can be harnessed to produce neural-trojan-style backdoors. To do so, we first outline a motivational attack scenario where a carefully chosen incorrect key, which we call a trojan key, produces misclassifications for an attacker-specified input class in a locked accelerator. We then develop a theoretically-robust attack methodology to automatically identify trojan keys. To evaluate this attack, we launch it on several locked accelerators. In our largest benchmark accelerator, our attack identified a trojan key that caused a 74\% decrease in classification accuracy for attacker-specified trigger inputs, while degrading accuracy by only 1.7\% for other inputs on average.

Runjia Zeng, Cheng Han, Qifan Wang et al.

With the scale of vision Transformer-based models continuing to grow, finetuning these large-scale pretrained models for new tasks has become increasingly parameter-intensive. Visual prompt tuning is introduced as a parameter-efficient finetuning (PEFT) method to this trend. Despite its successes, a notable research challenge persists within almost all PEFT approaches: significant performance degradation is observed when there is a substantial disparity between the datasets applied in pretraining and finetuning phases. To address this challenge, we draw inspiration from human visual cognition, and propose the Visual Fourier Prompt Tuning (VFPT) method as a general and effective solution for adapting large-scale transformer-based models. Our approach innovatively incorporates the Fast Fourier Transform into prompt embeddings and harmoniously considers both spatial and frequency domain information. Apart from its inherent simplicity and intuitiveness, VFPT exhibits superior performance across all datasets, offering a general solution to dataset challenges, irrespective of data disparities. Empirical results demonstrate that our approach outperforms current state-of-the-art baselines on two benchmarks, with low parameter usage (e.g., 0.57% of model parameters on VTAB-1k) and notable performance enhancements (e.g., 73.20% of mean accuracy on VTAB-1k). Our code is avaliable at https://github.com/runtsang/VFPT.

Runjia Zeng, Guangyan Sun, Qifan Wang et al.

Considering deep neural networks as manifold mappers, the pretrain-then-fine-tune paradigm can be interpreted as a two-stage process: pretrain establishes a broad knowledge base, and fine-tune adjusts the model parameters to activate specific neural pathways to align with the target manifold. Although prior fine-tuning approaches demonstrate success, their rigid parameter space limits their ability to dynamically activate appropriate neural pathways, rendering them ill-equipped to adapt flexibly to the diverse and evolving data distributions. In light of this view, we propose a novel approach, Mixture of Expert Prompt Tuning (MEPT), as an effective and efficient manifold-mapping framework. MEPT leverages the Mixture of Experts architecture by integrating multiple prompt experts to adaptively learn diverse and non-stationary data distributions. Empirical evaluations demonstrate that MEPT outperforms several state-of-the-art parameter efficient baselines on SuperGLUE, achieving notable improvements in mean accuracy (e.g., 1.94%) while significantly reducing activated prompts by 79.25%. The effectiveness of MEPT is further supported by theoretical insights from manifold learning and validated through neural activation pathway visualization results. Our code is avaliable at https://runjia.tech/emnlp_mept/.

Runjia Zeng, James Chenhao Liang, Cheng Han et al.

Training large language models (LLMs) from scratch can yield models with unique functionalities and strengths, but it is costly and often leads to redundant capabilities. A more cost-effective alternative is to fuse existing pre-trained LLMs with different architectures into a more powerful model. However, a key challenge in existing model fusion is their dependence on manually predefined vocabulary alignment, which may not generalize well across diverse contexts, leading to performance degradation in several evaluation. To solve this, we draw inspiration from distribution learning and propose the probabilistic token alignment method as a general and soft mapping for alignment, named as PTA-LLM. Our approach innovatively reformulates token alignment into a classic mathematical problem: optimal transport, seamlessly leveraging distribution-aware learning to facilitate more coherent model fusion. Apart from its inherent generality, PTA-LLM exhibits interpretability from a distributional perspective, offering insights into the essence of the token alignment. Empirical results demonstrate that probabilistic token alignment enhances the target model's performance across multiple capabilities. Our code is avaliable at https://runjia.tech/neurips_pta-llm/.

Li Yang, Qifan Wang, Jianfeng Chi et al.

Product attribute value extraction involves identifying the specific values associated with various attributes from a product profile. While existing methods often prioritize the development of effective models to improve extraction performance, there has been limited emphasis on extraction efficiency. However, in real-world scenarios, products are typically associated with multiple attributes, necessitating multiple extractions to obtain all corresponding values. In this work, we propose an Efficient product Attribute Value Extraction (EAVE) approach via lightweight sparse-layer interaction. Specifically, we employ a heavy encoder to separately encode the product context and attribute. The resulting non-interacting heavy representations of the context can be cached and reused for all attributes. Additionally, we introduce a light encoder to jointly encode the context and the attribute, facilitating lightweight interactions between them. To enrich the interaction within the lightweight encoder, we design a sparse-layer interaction module to fuse the non-interacting heavy representation into the lightweight encoder. Comprehensive evaluation on two benchmarks demonstrate that our method achieves significant efficiency gains with neutral or marginal loss in performance when the context is long and number of attributes is large. Our code is available \href{https://anonymous.4open.science/r/EAVE-EA18}{here}.

Liqi Yan, Yiming Cui, Yingjie Chen et al.

Geo-localization is a critical task in computer vision. In this work, we cast the geo-localization as a 2D image retrieval task. Current state-of-the-art methods for 2D geo-localization are not robust to locate a scene with drastic scale variations because they only exploit features from one semantic level for image representations. To address this limitation, we introduce a hierarchical attention fusion network using multi-scale features for geo-localization. We extract the hierarchical feature maps from a convolutional neural network (CNN) and organically fuse the extracted features for image representations. Our training is self-supervised using adaptive weights to control the attention of feature emphasis from each hierarchical level. Evaluation results on the image retrieval and the large-scale geo-localization benchmarks indicate that our method outperforms the existing state-of-the-art methods. Code is available here: \url{https://github.com/YanLiqi/HAF}.

Jiamian Wang, Guohao Sun, Pichao Wang et al.

The increasing prevalence of video clips has sparked growing interest in text-video retrieval. Recent advances focus on establishing a joint embedding space for text and video, relying on consistent embedding representations to compute similarity. However, the text content in existing datasets is generally short and concise, making it hard to fully describe the redundant semantics of a video. Correspondingly, a single text embedding may be less expressive to capture the video embedding and empower the retrieval. In this study, we propose a new stochastic text modeling method T-MASS, i.e., text is modeled as a stochastic embedding, to enrich text embedding with a flexible and resilient semantic range, yielding a text mass. To be specific, we introduce a similarity-aware radius module to adapt the scale of the text mass upon the given text-video pairs. Plus, we design and develop a support text regularization to further control the text mass during the training. The inference pipeline is also tailored to fully exploit the text mass for accurate retrieval. Empirical evidence suggests that T-MASS not only effectively attracts relevant text-video pairs while distancing irrelevant ones, but also enables the determination of precise text embeddings for relevant pairs. Our experimental results show a substantial improvement of T-MASS over baseline (3% to 6.3% by R@1). Also, T-MASS achieves state-of-the-art performance on five benchmark datasets, including MSRVTT, LSMDC, DiDeMo, VATEX, and Charades.

Cheng Han, Qifan Wang, Yiming Cui et al.

As the scale of vision models continues to grow, the emergence of Visual Prompt Tuning (VPT) as a parameter-efficient transfer learning technique has gained attention due to its superior performance compared to traditional full-finetuning. However, the conditions favoring VPT (the ``when") and the underlying rationale (the ``why") remain unclear. In this paper, we conduct a comprehensive analysis across 19 distinct datasets and tasks. To understand the ``when" aspect, we identify the scenarios where VPT proves favorable by two dimensions: task objectives and data distributions. We find that VPT is preferrable when there is 1) a substantial disparity between the original and the downstream task objectives (e.g., transitioning from classification to counting), or 2) a similarity in data distributions between the two tasks (e.g., both involve natural images). In exploring the ``why" dimension, our results indicate VPT's success cannot be attributed solely to overfitting and optimization considerations. The unique way VPT preserves original features and adds parameters appears to be a pivotal factor. Our study provides insights into VPT's mechanisms, and offers guidance for its optimal utilization.

Zhiyuan Cheng, Zhaoyi Liu, Tengda Guo et al.

Pixel-wise regression tasks (e.g., monocular depth estimation (MDE) and optical flow estimation (OFE)) have been widely involved in our daily life in applications like autonomous driving, augmented reality and video composition. Although certain applications are security-critical or bear societal significance, the adversarial robustness of such models are not sufficiently studied, especially in the black-box scenario. In this work, we introduce the first unified black-box adversarial patch attack framework against pixel-wise regression tasks, aiming to identify the vulnerabilities of these models under query-based black-box attacks. We propose a novel square-based adversarial patch optimization framework and employ probabilistic square sampling and score-based gradient estimation techniques to generate the patch effectively and efficiently, overcoming the scalability problem of previous black-box patch attacks. Our attack prototype, named BadPart, is evaluated on both MDE and OFE tasks, utilizing a total of 7 models. BadPart surpasses 3 baseline methods in terms of both attack performance and efficiency. We also apply BadPart on the Google online service for portrait depth estimation, causing 43.5% relative distance error with 50K queries. State-of-the-art (SOTA) countermeasures cannot defend our attack effectively.

Dong Hyun Jeon, Wenbo Sun, Houbing Herbert Song et al.

While deep-learning-enabled recommender systems demonstrate strong performance benchmarks, many struggle to adapt effectively in real-world environments due to limited use of user-item relationship data and insufficient transparency in recommendation generation. Traditional collaborative filtering approaches fail to integrate multifaceted item attributes, and although Factorization Machines account for item-specific details, they overlook broader relational patterns. Collaborative knowledge graph-based models have progressed by embedding user-item interactions with item-attribute relationships, offering a holistic perspective on interconnected entities. However, these models frequently aggregate attribute and interaction data in an implicit manner, leaving valuable relational nuances underutilized. This study introduces the Knowledge Graph Attention Network with Information Fusion (KGIF), a specialized framework designed to merge entity and relation embeddings explicitly through a tailored self-attention mechanism. The KGIF framework integrates reparameterization via dynamic projection vectors, enabling embeddings to adaptively represent intricate relationships within knowledge graphs. This explicit fusion enhances the interplay between user-item interactions and item-attribute relationships, providing a nuanced balance between user-centric and item-centric representations. An attentive propagation mechanism further optimizes knowledge graph embeddings, capturing multi-layered interaction patterns. The contributions of this work include an innovative method for explicit information fusion, improved robustness for sparse knowledge graphs, and the ability to generate explainable recommendations through interpretable path visualization.

Tingxu Han, Wei Song, Ziqi Ding et al.

Large language models (LLMs) increasingly mediate decisions in domains where unfair treatment of demographic groups is unacceptable. Existing work probes when biased outputs appear, but gives little insight into the mechanisms that generate them, leaving existing mitigations largely fragile. In this paper, we conduct a systematic investigation LLM unfairness and propose DiffHeads, a lightweight debiasing framework for LLMs. We first compare Direct-Answer (DA) prompting to Chain-of-Thought (CoT) prompting across eight representative open- and closed-source LLMs. DA will trigger the nature bias part of LLM and improve measured unfairness by 534.5%-391.9% in both one-turn and two-turn dialogues. Next, we define a token-to-head contribution score that traces each token's influence back to individual attention heads. This reveals a small cluster of bias heads that activate under DA but stay largely dormant with CoT, providing the first causal link between prompting strategy and bias emergence. Finally, building on this insight, we propose DiffHeads that identifies bias heads through differential activation analysis between DA and CoT, and selectively masks only those heads. DiffHeads reduces unfairness by 49.4%, and 40.3% under DA and CoT, respectively, without harming model utility.

Zhiyuan Cheng, Cheng Han, James Liang et al.

Monocular Depth Estimation (MDE) plays a vital role in applications such as autonomous driving. However, various attacks target MDE models, with physical attacks posing significant threats to system security. Traditional adversarial training methods, which require ground-truth labels, are not directly applicable to MDE models that lack ground-truth depth. Some self-supervised model hardening techniques (e.g., contrastive learning) overlook the domain knowledge of MDE, resulting in suboptimal performance. In this work, we introduce a novel self-supervised adversarial training approach for MDE models, leveraging view synthesis without the need for ground-truth depth. We enhance adversarial robustness against real-world attacks by incorporating L_0-norm-bounded perturbation during training. We evaluate our method against supervised learning-based and contrastive learning-based approaches specifically designed for MDE. Our experiments with two representative MDE networks demonstrate improved robustness against various adversarial attacks, with minimal impact on benign performance.

Yawen Lu, Dongfang Liu, Qifan Wang et al.

In this work, we introduce ProMotion, a unified prototypical framework engineered to model fundamental motion tasks. ProMotion offers a range of compelling attributes that set it apart from current task-specific paradigms. We adopt a prototypical perspective, establishing a unified paradigm that harmonizes disparate motion learning approaches. This novel paradigm streamlines the architectural design, enabling the simultaneous assimilation of diverse motion information. We capitalize on a dual mechanism involving the feature denoiser and the prototypical learner to decipher the intricacies of motion. This approach effectively circumvents the pitfalls of ambiguity in pixel-wise feature matching, significantly bolstering the robustness of motion representation. We demonstrate a profound degree of transferability across distinct motion patterns. This inherent versatility reverberates robustly across a comprehensive spectrum of both 2D and 3D downstream tasks. Empirical results demonstrate that ProMotion outperforms various well-known specialized architectures, achieving 0.54 and 0.054 Abs Rel error on the Sintel and KITTI depth datasets, 1.04 and 2.01 average endpoint error on the clean and final pass of Sintel flow benchmark, and 4.30 F1-all error on the KITTI flow benchmark. For its efficacy, we hope our work can catalyze a paradigm shift in universal models in computer vision.

Cheng Han, Yawen Lu, Guohao Sun et al.

In this work, we introduce the Prototypical Transformer (ProtoFormer), a general and unified framework that approaches various motion tasks from a prototype perspective. ProtoFormer seamlessly integrates prototype learning with Transformer by thoughtfully considering motion dynamics, introducing two innovative designs. First, Cross-Attention Prototyping discovers prototypes based on signature motion patterns, providing transparency in understanding motion scenes. Second, Latent Synchronization guides feature representation learning via prototypes, effectively mitigating the problem of motion uncertainty. Empirical results demonstrate that our approach achieves competitive performance on popular motion tasks such as optical flow and scene depth. Furthermore, it exhibits generality across various downstream tasks, including object tracking and video stabilization.

Cheng Han, James C. Liang, Qifan Wang et al.

We introduce the novel Diffusion Visual Programmer (DVP), a neuro-symbolic image translation framework. Our proposed DVP seamlessly embeds a condition-flexible diffusion model within the GPT architecture, orchestrating a coherent sequence of visual programs (i.e., computer vision models) for various pro-symbolic steps, which span RoI identification, style transfer, and position manipulation, facilitating transparent and controllable image translation processes. Extensive experiments demonstrate DVP's remarkable performance, surpassing concurrent arts. This success can be attributed to several key features of DVP: First, DVP achieves condition-flexible translation via instance normalization, enabling the model to eliminate sensitivity caused by the manual guidance and optimally focus on textual descriptions for high-quality content generation. Second, the framework enhances in-context reasoning by deciphering intricate high-dimensional concepts in feature spaces into more accessible low-dimensional symbols (e.g., [Prompt], [RoI object]), allowing for localized, context-free editing while maintaining overall coherence. Last but not least, DVP improves systemic controllability and explainability by offering explicit symbolic representations at each programming stage, empowering users to intuitively interpret and modify results. Our research marks a substantial step towards harmonizing artificial image translation processes with cognitive intelligence, promising broader applications.

James Liang, Tianfei Zhou, Dongfang Liu et al.

We present CLUSTSEG, a general, transformer-based framework that tackles different image segmentation tasks (i.e., superpixel, semantic, instance, and panoptic) through a unified neural clustering scheme. Regarding queries as cluster centers, CLUSTSEG is innovative in two aspects:1) cluster centers are initialized in heterogeneous ways so as to pointedly address task-specific demands (e.g., instance- or category-level distinctiveness), yet without modifying the architecture; and 2) pixel-cluster assignment, formalized in a cross-attention fashion, is alternated with cluster center update, yet without learning additional parameters. These innovations closely link CLUSTSEG to EM clustering and make it a transparent and powerful framework that yields superior results across the above segmentation tasks.

Qifan Wang, Yi Fang, Anirudh Ravula et al.

Structure information extraction refers to the task of extracting structured text fields from web pages, such as extracting a product offer from a shopping page including product title, description, brand and price. It is an important research topic which has been widely studied in document understanding and web search. Recent natural language models with sequence modeling have demonstrated state-of-the-art performance on web information extraction. However, effectively serializing tokens from unstructured web pages is challenging in practice due to a variety of web layout patterns. Limited work has focused on modeling the web layout for extracting the text fields. In this paper, we introduce WebFormer, a Web-page transFormer model for structure information extraction from web documents. First, we design HTML tokens for each DOM node in the HTML by embedding representations from their neighboring tokens through graph attention. Second, we construct rich attention patterns between HTML tokens and text tokens, which leverages the web layout for effective attention weight computation. We conduct an extensive set of experiments on SWDE and Common Crawl benchmarks. Experimental results demonstrate the superior performance of the proposed approach over several state-of-the-art methods.

Yiming Cui, Zhiwen Cao, Yixin Xie et al.

Multi-object tracking and segmentation (MOTS) is a critical task for autonomous driving applications. The existing MOTS studies face two critical challenges: 1) the published datasets inadequately capture the real-world complexity for network training to address various driving settings; 2) the working pipeline annotation tool is under-studied in the literature to improve the quality of MOTS learning examples. In this work, we introduce the DG-Labeler and DGL-MOTS dataset to facilitate the training data annotation for the MOTS task and accordingly improve network training accuracy and efficiency. DG-Labeler uses the novel Depth-Granularity Module to depict the instance spatial relations and produce fine-grained instance masks. Annotated by DG-Labeler, our DGL-MOTS dataset exceeds the prior effort (i.e., KITTI MOTS and BDD100K) in data diversity, annotation quality, and temporal representations. Results on extensive cross-dataset evaluations indicate significant performance improvements for several state-of-the-art methods trained on our DGL-MOTS dataset. We believe our DGL-MOTS Dataset and DG-Labeler hold the valuable potential to boost the visual perception of future transportation.

Yiming Cui, Liqi Yan, Zhiwen Cao et al.

Video objection detection is a challenging task because isolated video frames may encounter appearance deterioration, which introduces great confusion for detection. One of the popular solutions is to exploit the temporal information and enhance per-frame representation through aggregating features from neighboring frames. Despite achieving improvements in detection, existing methods focus on the selection of higher-level video frames for aggregation rather than modeling lower-level temporal relations to increase the feature representation. To address this limitation, we propose a novel solution named TF-Blender,which includes three modules: 1) Temporal relation mod-els the relations between the current frame and its neighboring frames to preserve spatial information. 2). Feature adjustment enriches the representation of every neigh-boring feature map; 3) Feature blender combines outputs from the first two modules and produces stronger features for the later detection tasks. For its simplicity, TF-Blender can be effortlessly plugged into any detection network to improve detection behavior. Extensive evaluations on ImageNet VID and YouTube-VIS benchmarks indicate the performance guarantees of using TF-Blender on recent state-of-the-art methods.

Dongfang Liu, Yiming Cui, Wenbo Tan et al.

Video instance segmentation (VIS) is a new and critical task in computer vision. To date, top-performing VIS methods extend the two-stage Mask R-CNN by adding a tracking branch, leaving plenty of room for improvement. In contrast, we approach the VIS task from a new perspective and propose a one-stage spatial granularity network (SG-Net). Compared to the conventional two-stage methods, SG-Net demonstrates four advantages: 1) Our method has a one-stage compact architecture and each task head (detection, segmentation, and tracking) is crafted interdependently so they can effectively share features and enjoy the joint optimization; 2) Our mask prediction is dynamically performed on the sub-regions of each detected instance, leading to high-quality masks of fine granularity; 3) Each of our task predictions avoids using expensive proposal-based RoI features, resulting in much reduced runtime complexity per instance; 4) Our tracking head models objects centerness movements for tracking, which effectively enhances the tracking robustness to different object appearances. In evaluation, we present state-of-the-art comparisons on the YouTube-VIS dataset. Extensive experiments demonstrate that our compact one-stage method can achieve improved performance in both accuracy and inference speed. We hope our SG-Net could serve as a strong and flexible baseline for the VIS task. Our code will be available.

Dongfang Liu, Yiming Cui, Liqi Yan et al.

In this work, we introduce a Denser Feature Network (DenserNet) for visual localization. Our work provides three principal contributions. First, we develop a convolutional neural network (CNN) architecture which aggregates feature maps at different semantic levels for image representations. Using denser feature maps, our method can produce more keypoint features and increase image retrieval accuracy. Second, our model is trained end-to-end without pixel-level annotation other than positive and negative GPS-tagged image pairs. We use a weakly supervised triplet ranking loss to learn discriminative features and encourage keypoint feature repeatability for image representation. Finally, our method is computationally efficient as our architecture has shared features and parameters during computation. Our method can perform accurate large-scale localization under challenging conditions while remaining the computational constraint. Extensive experiment results indicate that our method sets a new state-of-the-art on four challenging large-scale localization benchmarks and three image retrieval benchmarks.

Zhiwen Cao, Zongcheng Chu, Dongfang Liu et al.

This paper proposes to use the three vectors in a rotation matrix as the representation in head pose estimation and develops a new neural network based on the characteristic of such representation. We address two potential issues existed in current head pose estimation works: 1. Public datasets for head pose estimation use either Euler angles or quaternions to annotate data samples. However, both of these annotations have the issue of discontinuity and thus could result in some performance issues in neural network training. 2. Most research works report Mean Absolute Error (MAE) of Euler angles as the measurement of performance. We show that MAE may not reflect the actual behavior especially for the cases of profile views. To solve these two problems, we propose a new annotation method which uses three vectors to describe head poses and a new measurement Mean Absolute Error of Vectors (MAEV) to assess the performance. We also train a new neural network to predict the three vectors with the constraints of orthogonality. Our proposed method achieves state-of-the-art results on both AFLW2000 and BIWI datasets. Experiments show our vector-based annotation method can effectively reduce prediction errors for large pose angles.

Liqi Yan, Dongfang Liu, Yaoxian Song et al.

Vision and voice are two vital keys for agents' interaction and learning. In this paper, we present a novel indoor navigation model called Memory Vision-Voice Indoor Navigation (MVV-IN), which receives voice commands and analyzes multimodal information of visual observation in order to enhance robots' environment understanding. We make use of single RGB images taken by a first-view monocular camera. We also apply a self-attention mechanism to keep the agent focusing on key areas. Memory is important for the agent to avoid repeating certain tasks unnecessarily and in order for it to adapt adequately to new scenes, therefore, we make use of meta-learning. We have experimented with various functional features extracted from visual observation. Comparative experiments prove that our methods outperform state-of-the-art baselines.

Dongfang Liu, Yiming Cui, Xiaolei Guo et al.

Accurate localization is a foundational capacity, required for autonomous vehicles to accomplish other tasks such as navigation or path planning. It is a common practice for vehicles to use GPS to acquire location information. However, the application of GPS can result in severe challenges when vehicles run within the inner city where different kinds of structures may shadow the GPS signal and lead to inaccurate location results. To address the localization challenges of urban settings, we propose a novel feature voting technique for visual localization. Different from the conventional front-view-based method, our approach employs views from three directions (front, left, and right) and thus significantly improves the robustness of location prediction. In our work, we craft the proposed feature voting method into three state-of-the-art visual localization networks and modify their architectures properly so that they can be applied for vehicular operation. Extensive field test results indicate that our approach can predict location robustly even in challenging inner-city settings. Our research sheds light on using the visual localization approach to help autonomous vehicles to find accurate location information in a city maze, within a desirable time constraint.