Mobile agents show immense potential, yet current state-of-the-art (SoTA) agents exhibit inadequate success rates on real-world, long-horizon, cross-application tasks. We attribute this bottleneck to the agents' excessive reliance on static, internal knowledge within MLLMs, which leads to two critical failure points: 1) strategic hallucinations in high-level planning and 2) operational errors during low-level execution on user interfaces (UI). The core insight of this paper is that high-level planning and low-level UI operations require fundamentally distinct types of knowledge. Planning demands high-level, strategy-oriented experiences, whereas operations necessitate low-level, precise instructions closely tied to specific app UIs. Motivated by these insights, we propose Mobile-Agent-RAG, a novel hierarchical multi-agent framework that innovatively integrates dual-level retrieval augmentation. At the planning stage, we introduce Manager-RAG to reduce strategic hallucinations by retrieving human-validated comprehensive task plans that provide high-level guidance. At the execution stage, we develop Operator-RAG to improve execution accuracy by retrieving the most precise low-level guidance for accurate atomic actions, aligned with the current app and subtask. To accurately deliver these knowledge types, we construct two specialized retrieval-oriented knowledge bases. Furthermore, we introduce Mobile-Eval-RAG, a challenging benchmark for evaluating such agents on realistic multi-app, long-horizon tasks. Extensive experiments demonstrate that Mobile-Agent-RAG significantly outperforms SoTA baselines, improving task completion rate by 11.0% and step efficiency by 10.2%, establishing a robust paradigm for context-aware, reliable multi-agent mobile automation.

In this paper, we propose FakeRadar, a novel deepfake video detection framework designed to address the challenges of cross-domain generalization in real-world scenarios. Existing detection methods typically rely on manipulation-specific cues, performing well on known forgery types but exhibiting severe limitations against emerging manipulation techniques. This poor generalization stems from their inability to adapt effectively to unseen forgery patterns. To overcome this, we leverage large-scale pretrained models (e.g. CLIP) to proactively probe the feature space, explicitly highlighting distributional gaps between real videos, known forgeries, and unseen manipulations. Specifically, FakeRadar introduces Forgery Outlier Probing, which employs dynamic subcluster modeling and cluster-conditional outlier generation to synthesize outlier samples near boundaries of estimated subclusters, simulating novel forgery artifacts beyond known manipulation types. Additionally, we design Outlier-Guided Tri-Training, which optimizes the detector to distinguish real, fake, and outlier samples using proposed outlier-driven contrastive learning and outlier-conditioned cross-entropy losses. Experiments show that FakeRadar outperforms existing methods across various benchmark datasets for deepfake video detection, particularly in cross-domain evaluations, by handling the variety of emerging manipulation techniques.

This paper aims to restore original background images in watermarked videos, overcoming challenges posed by traditional approaches that fail to handle the temporal dynamics and diverse watermark characteristics effectively. Our method introduces a unique framework that first "decouples" the extraction of prior knowledge-such as common-sense knowledge and residual background details-from the temporal modeling process, allowing for independent handling of background restoration and temporal consistency. Subsequently, it "couples" these extracted features by integrating them into the temporal modeling backbone of a video inpainting (VI) framework. This integration is facilitated by a specialized module, which includes an intrinsic background image prediction sub-module and a dual-branch frame embedding module, designed to reduce watermark interference and enhance the application of prior knowledge. Moreover, a frame-adaptive feature selection module dynamically adjusts the extraction of prior features based on the corruption level of each frame, ensuring their effective incorporation into the temporal processing. Extensive experiments on YouTube-VOS and DAVIS datasets validate our method's efficiency in watermark removal and background restoration, showing significant improvement over state-of-the-art techniques in visible image watermark removal, video restoration, and video inpainting.

Federated Learning with Noisy Labels (F-LNL) aims at seeking an optimal server model via collaborative distributed learning by aggregating multiple client models trained with local noisy or clean samples. On the basis of a federated learning framework, recent advances primarily adopt label noise filtering to separate clean samples from noisy ones on each client, thereby mitigating the negative impact of label noise. However, these prior methods do not learn noise filters by exploiting knowledge across all clients, leading to sub-optimal and inferior noise filtering performance and thus damaging training stability. In this paper, we present FedDiv to tackle the challenges of F-LNL. Specifically, we propose a global noise filter called Federated Noise Filter for effectively identifying samples with noisy labels on every client, thereby raising stability during local training sessions. Without sacrificing data privacy, this is achieved by modeling the global distribution of label noise across all clients. Then, in an effort to make the global model achieve higher performance, we introduce a Predictive Consistency based Sampler to identify more credible local data for local model training, thus preventing noise memorization and further boosting the training stability. Extensive experiments on CIFAR-10, CIFAR-100, and Clothing1M demonstrate that FedDiv achieves superior performance over state-of-the-art F-LNL methods under different label noise settings for both IID and non-IID data partitions. Source code is publicly available at https://github.com/lijichang/FLNL-FedDiv.

Powered by massive curated training data, Segment Any-thing Model (SAM) has demonstrated its impressive generalization capabilities in open-world scenarios with the guidance of prompts. However, the vanilla SAM is class-agnostic and heavily relies on user-provided prompts to segment objects of interest. Adapting this method to diverse tasks is crucial for accurate target identification and to avoid suboptimal segmentation results. In this paper, we propose a novel framework, termed AlignSAM, designed for automatic prompting for aligning SAM to an open context through reinforcement learning. Anchored by an agent, AlignSAM enables the generality of the SAM model across diverse downstream tasks while keeping its parameters frozen. Specifically, AlignSAM initiates a prompting agent to iteratively refine segmentation predictions by interacting with the foundational model. It integrates a reinforcement learning policy network to provide informative prompts to the foundational models. Additionally, a semantic recal-ibration module is introduced to provide fine-grained labels of prompts, enhancing the model's proficiency in handling tasks encompassing explicit and implicit semantics. Experiments conducted on various challenging segmentation tasks among existing foundation models demonstrate the superiority of the proposed AlignSAM over state-of-the-art approaches. Project page: https://github.com/Duojun-Huang/AIignSAM-CVPR2024.

Open vocabulary object detection (OVD) aims at seeking an optimal object detector capable of recognizing objects from both base and novel categories. Recent advances leverage knowledge distillation to transfer insightful knowledge from pre-trained large-scale vision-language models to the task of object detection, significantly generalizing the powerful capabilities of the detector to identify more unknown object categories. However, these methods face significant challenges in background interpretation and model overfitting and thus often result in the loss of crucial back-ground knowledge, giving rise to sub-optimal inference performance of the detector. To mitigate these issues, we present a novel OVD framework termed LBP to propose learning background prompts to harness explored implicit background knowledge, thus enhancing the detection performance w.r.t. base and novel categories. Specifically, we devise three modules: Background Category-specific Prompt, Background Object Discovery, and Inference Probability Rectification, to empower the detector to discover, represent, and leverage implicit object knowledge explored from background proposals. Evaluation on two benchmark datasets, OV-COCO and OV-LVIS, demonstrates the superiority of our proposed method over existing state-of-the-art approaches in handling the OVD tasks.

Active domain adaptation (ADA) aims to improve the model adaptation performance by incorporating active learning (AL) techniques to label a maximally-informative subset of target samples. Conventional AL methods do not consider the existence of domain shift, and hence, fail to identify the truly valuable samples in the context of domain adaptation. To accommodate active learning and domain adaption, the two naturally different tasks, in a collaborative framework, we advocate that a customized learning strategy for the target data is the key to the success of ADA solutions. We present Divide-and-Adapt (DiaNA), a new ADA framework that partitions the target instances into four categories with stratified transferable properties. With a novel data subdivision protocol based on uncertainty and domainness, DiaNA can accurately recognize the most gainful samples. While sending the informative instances for annotation, DiaNA employs tailored learning strategies for the remaining categories. Furthermore, we propose an informativeness score that unifies the data partitioning criteria. This enables the use of a Gaussian mixture model (GMM) to automatically sample unlabeled data into the proposed four categories. Thanks to the “divide-and-adapt” spirit, DiaNA can handle data with large variations of domain gap. In addition, we show that DiaNA can generalize to different domain adaptation settings, such as unsupervised domain adaptation (UDA), semi-supervised domain adaptation (SSDA), source-free domain adaptation (SFDA), etc.

Compared to unsupervised domain adaptation, semi-supervised domain adaptation (SSDA) aims to significantly improve the classification performance and generalization capability of the model by leveraging the presence of a small amount of labeled data from the target domain. Several SSDA approaches have been developed to enable semantic-aligned feature confusion between labeled (or pseudo labeled) samples across domains; nevertheless, owing to the scarcity of semantic label information of the target domain, they were arduous to fully realize their potential. In this study, we propose a novel SSDA approach named Graph-based Adaptive Betweenness Clustering (G-ABC) for achieving categorical domain alignment, which enables cross-domain semantic alignment by mandating semantic transfer from labeled data of both the source and target domains to unlabeled target samples. In particular, a heterogeneous graph is initially constructed to reflect the pairwise relationships between labeled samples from both domains and unlabeled ones of the target domain. Then, to degrade the noisy connectivity in the graph, connectivity refinement is conducted by introducing two strategies, namely Confidence Uncertainty based Node Removal and Prediction Dissimilarity based Edge Pruning. Once the graph has been refined, Adaptive Betweenness Clustering is introduced to facilitate semantic transfer by using across-domain betweenness clustering and within-domain betweenness clustering, thereby propagating semantic label information from labeled samples across domains to unlabeled target data. Extensive experiments on three standard benchmark datasets, namely DomainNet, Office-Home, and Office-31, indicated that our method outperforms previous state-of-the-art SSDA approaches, demonstrating the superiority of the proposed G-ABC algorithm.

In semi-supervised domain adaptation, a few labeled samples per class in the target domain guide features of the remaining target samples to aggregate around them. However, the trained model cannot produce a highly discriminative feature representation for the target domain because the training data is dominated by labeled samples from the source domain. This could lead to disconnection between the labeled and unlabeled target samples as well as misalignment between unlabeled target samples and the source domain. In this paper, we propose a novel approach called Cross-domain Adaptive Clustering to address this problem. To achieve both inter-domain and intra-domain adaptation, we first introduce an adversarial adaptive clustering loss to group features of unlabeled target data into clusters and perform cluster-wise feature alignment across the source and target domains. We further apply pseudo labeling to unlabeled samples in the target domain and retain pseudo-labels with high confidence. Pseudo labeling expands the number of "labeled" samples in each class in the target domain, and thus produces a more robust and powerful cluster core for each class to facilitate adversarial learning. Extensive experiments on benchmark datasets, including DomainNet, Office-Home and Office, demonstrate that our proposed approach achieves the state-of-the-art performance in semi-supervised domain adaptation.

Using an ensemble of neural networks with consistency regularization is effective for improving performance and stability of deep learning, compared to the case of a single network. In this paper, we present a semi-supervised Deep Coupled Ensemble (DCE) model, which contributes to ensemble learning and classification landmark exploration for better locating the final decision boundaries in the learnt latent space. First, multiple complementary consistency regularizations are integrated into our DCE model to enable the ensemble members to learn from each other and themselves, such that training experience from different sources can be shared and utilized during training. Second, in view of the possibility of producing incorrect predictions on a number of difficult instances, we adopt class-wise mean feature matching to explore important unlabeled instances as classification landmarks, on which the model predictions are more reliable. Minimizing the weighted conditional entropy on unlabeled data is able to force the final decision boundaries to move away from important training data points, which facilitates semi-supervised learning. Ensemble members could eventually have similar performance due to consistency regularization, and thus only one of these members is needed during the test stage, such that the efficiency of our model is the same as the non-ensemble case. Extensive experimental results demonstrate the superiority of our proposed DCE model over existing state-of-the-art semi-supervised learning methods.

Learning class-conditional data distributions is crucial for Generative Adversarial Networks (GAN) in semi-supervised learning. To improve both instance synthesis and classification in this setting, we propose an enhanced TripleGAN (EnhancedTGAN) model in this work. We follow the adversarial training scheme of the original TripleGAN, but completely re-design the training targets of the generator and classifier. Specifically, we adopt feature-semantics matching to enhance the generator in learning class-conditional distributions from both the aspects of statistics in the latent space and semantics consistency with respect to the generator and classifier. Since a limited amount of labeled data is not sufficient to determine satisfactory decision boundaries, we include two classifiers, and incorporate collaborative learning into our model to provide better guidance for generator training. The synthesized high-fidelity data can in turn be used for improving classifier training. In the experiments, the superior performance of our approach on multiple benchmark datasets demonstrates the effectiveness of the mutual reinforcement between the generator and classifiers in facilitating semi-supervised instance synthesis and classification.

Deep mutual learning jointly trains multiple essential networks having similar properties to improve semi-supervised classification. However, the commonly used consistency regularization between the outputs of the networks may not fully leverage the difference between them. In this paper, we explore how to capture the complementary information to enhance mutual learning. For this purpose, we propose a complementary correction network (CCN), built on top of the essential networks, to learn the mapping from the output of one essential network to the ground truth label, conditioned on the features learnt by another. To make the second essential network increasingly complementary to the first one, this network is supervised by the corrected predictions. As a result, minimizing the prediction divergence between the two complementary networks can lead to significant performance gains in semi-supervised learning. Our experimental results demonstrate that the proposed approach clearly improves mutual learning between essential networks, and achieves state-of-the-art results on multiple semi-supervised classification benchmarks. In particular, the test error rates are reduced from previous 21.23% and 14.65% to 12.05% and 10.37% on CIFAR-10 with 1000 and 2000 labels, respectively.

Previous studies have shown that attention mechanisms and shortest dependency paths have a positive effect on relation classification. In this paper, a keyword-attentive sentence mechanism is proposed to effectively combine the two methods. Furthermore, to effectively handle the imbalanced classification problem, this paper proposes a new loss function called the synthetic stimulation loss, which uses a modulating factor to allow the model to focus on hard-to-classify samples. The proposed two methods are integrated into a bidirectional gated recurrent unit (BiGRU). As a single model is not strong in noise immunity, this paper applies the mutual learning method to our model and forces the networks to teach each other. Therefore, we call the final model SSL-KAS-MuBiGRU. Experiments on the SemEval-2010 Task 8 data set and the TAC40 data set demonstrate that the keyword-attentive sentence mechanism and synthetic stimulation loss are useful for relation classification, and our model achieves state-of-the-art results.