Zhiheng Ma

Hui Lin, Zhiheng Ma, Rongrong Ji et al.

This paper focuses on the challenging crowd counting task. As large-scale variations often exist within crowd images, neither fixed-size convolution kernel of CNN nor fixed-size attention of recent vision transformers can well handle this kind of variation. To address this problem, we propose a Multifaceted Attention Network (MAN) to improve transformer models in local spatial relation encoding. MAN incorporates global attention from a vanilla transformer, learnable local attention, and instance attention into a counting model. Firstly, the local Learnable Region Attention (LRA) is proposed to assign attention exclusively for each feature location dynamically. Secondly, we design the Local Attention Regularization to supervise the training of LRA by minimizing the deviation among the attention for different feature locations. Finally, we provide an Instance Attention mechanism to focus on the most important instances dynamically during training. Extensive experiments on four challenging crowd counting datasets namely ShanghaiTech, UCF-QNRF, JHU++, and NWPU have validated the proposed method. Codes: https://github.com/LoraLinH/Boosting-Crowd-Counting-via-Multifaceted-Attention.

Yabin Wang, Zhiheng Ma, Zhiwu Huang et al.

This paper focuses on the prevalent performance imbalance in the stages of incremental learning. To avoid obvious stage learning bottlenecks, we propose a brand-new stage-isolation based incremental learning framework, which leverages a series of stage-isolated classifiers to perform the learning task of each stage without the interference of others. To be concrete, to aggregate multiple stage classifiers as a uniform one impartially, we first introduce a temperature-controlled energy metric for indicating the confidence score levels of the stage classifiers. We then propose an anchor-based energy self-normalization strategy to ensure the stage classifiers work at the same energy level. Finally, we design a voting-based inference augmentation strategy for robust inference. The proposed method is rehearsal free and can work for almost all continual learning scenarios. We evaluate the proposed method on four large benchmarks. Extensive results demonstrate the superiority of the proposed method in setting up new state-of-the-art overall performance. \emph{Code is available at} \url{https://github.com/iamwangyabin/ESN}.

Haoyun Liu, Jianzhuang Zhao, Xinyuan Chang et al.

Despite the rapid progress of vision-language-action (VLA) models, the prevailing practice of predicting action chunks as discrete waypoints remains structurally misaligned with the intrinsic continuity of physical motion. This discretization arises naturally from fixed-rate robot data collection and the token-by-token prediction paradigm of large language models, but ties actions to rigid sampling rates, does not naturally support analytically consistent higher-order derivatives, and introduces quantization artifacts that hinder precise, compliant interaction. We propose Neural Implicit Action Fields (NIAF), which reformulates chunk-level action representation from discrete waypoints to continuous action functions. Using a vision-language model as a hierarchical spectral modulator over a learnable motion prior, NIAF synthesizes continuous-time action manifolds with arbitrary temporal resolution. This formulation enables analytical differentiation, allowing explicit supervision of velocity and regularization of higher-order derivative signals to promote mathematical consistency, physical plausibility, and control smoothness. Our approach achieves strong results on CALVIN and LIBERO across diverse backbones. Real-world experiments further confirm that NIAF supports stable impedance control, bridging policy-side action generation and execution-side smooth control.

Hui Lin, Zhiheng Ma, Xiaopeng Hong et al.

In this paper, we propose a new agency-guided semi-supervised counting approach. First, we build a learnable auxiliary structure, namely the density agency to bring the recognized foreground regional features close to corresponding density sub-classes (agents) and push away background ones. Second, we propose a density-guided contrastive learning loss to consolidate the backbone feature extractor. Third, we build a regression head by using a transformer structure to refine the foreground features further. Finally, an efficient noise depression loss is provided to minimize the negative influence of annotation noises. Extensive experiments on four challenging crowd counting datasets demonstrate that our method achieves superior performance to the state-of-the-art semi-supervised counting methods by a large margin. Code is available.

Zhiheng Ma, Xiaopeng Hong, Qinnan Shangguan

Meta AI recently released the Segment Anything model (SAM), which has garnered attention due to its impressive performance in class-agnostic segmenting. In this study, we explore the use of SAM for the challenging task of few-shot object counting, which involves counting objects of an unseen category by providing a few bounding boxes of examples. We compare SAM's performance with other few-shot counting methods and find that it is currently unsatisfactory without further fine-tuning, particularly for small and crowded objects. Code can be found at \url{https://github.com/Vision-Intelligence-and-Robots-Group/count-anything}.

Zhiheng Ma, Xiaopeng Hong, Beinan Liu et al.

Although data-free incremental learning methods are memory-friendly, accurately estimating and counteracting representation shifts is challenging in the absence of historical data. This paper addresses this thorny problem by proposing a novel incremental learning method inspired by human analogy capabilities. Specifically, we design an analogy-making mechanism to remap the new data into the old class by prompt tuning. It mimics the feature distribution of the target old class on the old model using only samples of new classes. The learnt prompts are further used to estimate and counteract the representation shift caused by fine-tuning for the historical prototypes. The proposed method sets up new state-of-the-art performance on four incremental learning benchmarks under both the class and domain incremental learning settings. It consistently outperforms data-replay methods by only saving feature prototypes for each class. It has almost hit the empirical upper bound by joint training on the Core50 benchmark. The code will be released at \url{https://github.com/ZhihengCV/A-Prompts}.

Yabin Wang, Xiaopeng Hong, Zhiheng Ma et al.

Task allocation plays a vital role in multi-robot autonomous cleaning systems, where multiple robots work together to clean a large area. However, most current studies mainly focus on deterministic, single-task allocation for cleaning robots, without considering hybrid tasks in uncertain working environments. Moreover, there is a lack of datasets and benchmarks for relevant research. In this paper, to address these problems, we formulate multi-robot hybrid-task allocation under the uncertain cleaning environment as a robust optimization problem. Firstly, we propose a novel robust mixed-integer linear programming model with practical constraints including the task order constraint for different tasks and the ability constraints of hybrid robots. Secondly, we establish a dataset of \emph{100} instances made from floor plans, each of which has 2D manually-labeled images and a 3D model. Thirdly, we provide comprehensive results on the collected dataset using three traditional optimization approaches and a deep reinforcement learning-based solver. The evaluation results show that our solution meets the needs of multi-robot cleaning task allocation and the robust solver can protect the system from worst-case scenarios with little additional cost. The benchmark will be available at {https://github.com/iamwangyabin/Multi-robot-Cleaning-Task-Allocation}.

Zhilin Zhu, Xiaopeng Hong, Zhiheng Ma et al.

Continual Test-Time Adaptation (CTTA) involves adapting a pre-trained source model to continually changing unsupervised target domains. In this paper, we systematically analyze the challenges of this task: online environment, unsupervised nature, and the risks of error accumulation and catastrophic forgetting under continual domain shifts. To address these challenges, we reshape the online data buffering and organizing mechanism for CTTA. We propose an uncertainty-aware buffering approach to identify and aggregate significant samples with high certainty from the unsupervised, single-pass data stream. Based on this, we propose a graph-based class relation preservation constraint to overcome catastrophic forgetting. Furthermore, a pseudo-target replay objective is used to mitigate error accumulation. Extensive experiments demonstrate the superiority of our method in both segmentation and classification CTTA tasks. Code is available at https://github.com/z1358/OBAO.

Chenhao Wang, Xiaopeng Hong, Zhiheng Ma et al.

Multi-modal crowd counting is a crucial task that uses multi-modal cues to estimate the number of people in crowded scenes. To overcome the gap between different modalities, we propose a modal emulation-based two-pass multi-modal crowd-counting framework that enables efficient modal emulation, alignment, and fusion. The framework consists of two key components: a \emph{multi-modal inference} pass and a \emph{cross-modal emulation} pass. The former utilizes a hybrid cross-modal attention module to extract global and local information and achieve efficient multi-modal fusion. The latter uses attention prompting to coordinate different modalities and enhance multi-modal alignment. We also introduce a modality alignment module that uses an efficient modal consistency loss to align the outputs of the two passes and bridge the semantic gap between modalities. Extensive experiments on both RGB-Thermal and RGB-Depth counting datasets demonstrate its superior performance compared to previous methods. Code available at https://github.com/Mr-Monday/Multi-modal-Crowd-Counting-via-Modal-Emulation.

Zhilin Zhu, Yabin Wang, Zhiheng Ma et al.

Continual Test-Time Adaptation (CTTA) aims to empower perception systems to handle dynamic distribution shifts encountered after deployment. Existing methods predominantly follow a backward-alignment paradigm, which rigidly aligns incoming data with supervisory surrogates derived from the source domain. Consequently, they struggle with unreliable supervision and evolving distribution shifts. To overcome these limitations, we introduce a novel forward-facilitation paradigm through a method termed Dynamic Style Bridging. Prior to deployment, we construct a compact knowledge base of generated class exemplars. During test time, to mitigate inherent generative bias and adapt these proxies to incoming data, we propose a multi-level bridging mechanism. This mechanism dynamically injects the proxies with incoming data styles at the input, statistical, and representation levels, while preserving the original semantics of the proxies. These high-fidelity proxies are then used to provide reliable, on-demand supervisory signals, enabling stable adaptation under continual shifts. Extensive experiments across standard CTTA benchmarks demonstrate that our method achieves consistent and substantial improvements over recent state-of-the-art approaches. Code is available at \href{https://github.com/z1358/DAS}.

Jiangyang Li, Cong Wan, Changjie Wu et al.

Reliable spatial reasoning remains a core bottleneck for vision-language models (VLMs). Existing mainstream training paradigms for spatial reasoning largely rely on outcome alignment or process imitation, lacking explicit constraints on the reasoning process, and therefore struggle to ensure genuine visual dependence and stable reasoning trajectories. In this paper, we construct a high-quality CoT dataset covering diverse spatial phenomena and diagnose the model's reasoning process, revealing two typical types of process degradation during reinforcement learning optimization: Spurious Grounding, which bypasses visual evidence, and Tail Instability, where uncertainty abnormally rises in the later stage of reasoning. To address these issues, we propose ProSR, a process-shaping optimization framework for spatial reasoning. Through a Counterfactual Invariance Penalty and a Tail Drift Penalty, ProSR extends the optimization objective from single answer correctness to two process-level dimensions: visual dependence and trajectory stability. Experiments on multiple complex and out-of-distribution spatial reasoning benchmarks show that ProSR improves answer accuracy while generating reasoning trajectories that are more stable and more dependent on visual evidence.

Dongjie Huo, Haoyun Liu, Guoqing Liu et al.

Current embodied intelligent systems still face a substantial gap between high-level reasoning and low-level physical execution in open-world environments. Although Vision-Language-Action (VLA) models provide strong perception and intuitive responses, their open-loop nature limits long-horizon performance. Agents incorporating System 2 cognitive mechanisms improve planning, but usually operate in closed sandboxes with predefined toolkits and limited real-system control. OpenClaw provides a localized runtime with full system privileges, but lacks the embodied control architecture required for long-duration, multi-robot execution. We therefore propose ABot-Claw, an embodied extension of OpenClaw that integrates: 1) a unified embodiment interface with capability-driven scheduling for heterogeneous robot coordination; 2) a visual-centric cross-embodiment multimodal memory for persistent context retention and grounded retrieval; and 3) a critic-based closed-loop feedback mechanism with a generalist reward model for online progress evaluation, local correction, and replanning. With a decoupled architecture spanning the OpenClaw layer, shared service layer, and robot embodiment layer, ABot-Claw enables real-world interaction, closes the loop from natural language intent to physical action, and supports progressively self-evolving robotic agents in open, dynamic environments.

Hui Lin, Zhiheng Ma, Xiaopeng Hong et al.

Transformer has been popular in recent crowd counting work since it breaks the limited receptive field of traditional CNNs. However, since crowd images always contain a large number of similar patches, the self-attention mechanism in Transformer tends to find a homogenized solution where the attention maps of almost all patches are identical. In this paper, we address this problem by proposing Gramformer: a graph-modulated transformer to enhance the network by adjusting the attention and input node features respectively on the basis of two different types of graphs. Firstly, an attention graph is proposed to diverse attention maps to attend to complementary information. The graph is building upon the dissimilarities between patches, modulating the attention in an anti-similarity fashion. Secondly, a feature-based centrality encoding is proposed to discover the centrality positions or importance of nodes. We encode them with a proposed centrality indices scheme to modulate the node features and similarity relationships. Extensive experiments on four challenging crowd counting datasets have validated the competitiveness of the proposed method. Code is available at {https://github.com/LoraLinH/Gramformer}.

Jiangyang Li, Cong Wan, SongLin Dong et al.

Vision-and-Language Navigation (VLN) requires agents to navigate photo-realistic environments following natural language instructions. Current methods predominantly rely on imitation learning, which suffers from limited generalization and poor robustness to execution perturbations. We present NavGRPO, a reinforcement learning framework that learns goal-directed navigation policies through Group Relative Policy Optimization. By exploring diverse trajectories and optimizing via within-group performance comparisons, our method enables agents to distinguish effective strategies beyond expert paths without requiring additional value networks. Built on ScaleVLN, NavGRPO achieves superior robustness on R2R and REVERIE benchmarks with +3.0% and +1.71% SPL improvements in unseen environments. Under extreme early-stage perturbations, we demonstrate +14.89% SPL gain over the baseline, confirming that goal-directed RL training builds substantially more robust navigation policies. Code and models will be released.

Zhiheng Ma, Anjia Cao, Funing Yang et al.

Most dataset distillation methods struggle to accommodate large-scale datasets due to their substantial computational and memory requirements. Recent research has begun to explore scalable disentanglement methods. However, there are still performance bottlenecks and room for optimization in this direction. In this paper, we present a curriculum-based dataset distillation framework aiming to harmonize performance and scalability. This framework strategically distills synthetic images, adhering to a curriculum that transitions from simple to complex. By incorporating curriculum evaluation, we address the issue of previous methods generating images that tend to be homogeneous and simplistic, doing so at a manageable computational cost. Furthermore, we introduce adversarial optimization towards synthetic images to further improve their representativeness and safeguard against their overfitting to the neural network involved in distilling. This enhances the generalization capability of the distilled images across various neural network architectures and also increases their robustness to noise. Extensive experiments demonstrate that our framework sets new benchmarks in large-scale dataset distillation, achieving substantial improvements of 11.1\% on Tiny-ImageNet, 9.0\% on ImageNet-1K, and 7.3\% on ImageNet-21K. Our distilled datasets and code are available at https://github.com/MIV-XJTU/CUDD.

Yuzhi Chen, Ronghan Chen, Dongjie Huo et al.

Video-based world models offer a powerful paradigm for embodied simulation and planning, yet state-of-the-art models often generate physically implausible manipulations - such as object penetration and anti-gravity motion - due to training on generic visual data and likelihood-based objectives that ignore physical laws. We present ABot-PhysWorld, a 14B Diffusion Transformer model that generates visually realistic, physically plausible, and action-controllable videos. Built on a curated dataset of three million manipulation clips with physics-aware annotation, it uses a novel DPO-based post-training framework with decoupled discriminators to suppress unphysical behaviors while preserving visual quality. A parallel context block enables precise spatial action injection for cross-embodiment control. To better evaluate generalization, we introduce EZSbench, the first training-independent embodied zero-shot benchmark combining real and synthetic unseen robot-task-scene combinations. It employs a decoupled protocol to separately assess physical realism and action alignment. ABot-PhysWorld achieves new state-of-the-art performance on PBench and EZSbench, surpassing Veo 3.1 and Sora v2 Pro in physical plausibility and trajectory consistency. We will release EZSbench to promote standardized evaluation in embodied video generation.

Detao Bai, Shimin Yao, Weixuan Chen et al.

Recent advances in omni-modal large language models have enabled remarkable progress in joint vision-audio understanding. However, prevailing architectures rely on modality-specific encoders with a \emph{video-coarse, audio-dense} design -- sampling visual frames at 1--2 fps while processing audio waveforms at 25 fps -- resulting in systems that perceive video \emph{frame by frame, modality by modality} rather than holistically as humans do. Such a discrepancy leaves models with impoverished cross-modal interaction during encoding and an inability to capture fine-grained visual motion. To bridge this gap, we present \textbf{Omni-Encoder, a unified Transformer backbone designed to co-embed visual and audio signals at a symmetrical 25 fps} within a shared latent space. This architecture leverages three core innovations -- the Omni-Encoder Token Template, Omni-RoPE, and Temporal Window Shifting -- to effectively reconcile the dual challenges of modality disentanglement and computational efficiency. Experiments demonstrate that, compared to the modality-specific baseline Qwen2.5-Omni under the same input token budget to the LLM decoder, Omni-Encoder delivers substantial gains on visual continuous understanding tasks -- such as sign language recognition and fine-grained sports action analysis -- while maintaining competitive performance on established audio-visual benchmarks such as AVQA and Speaker Identification and Localization. These results suggest that unified omnivorous encoding offers a promising direction for building omni-modal models that more closely reflect the integrated nature of human perception.

Yaohui Ma, Xiaopeng Hong, Shizhou Zhang et al.

Large multimodal language models (MLLMs) have revolutionized natural language processing and visual understanding, but often contain outdated or inaccurate information. Current multimodal knowledge editing evaluations are limited in scope and potentially biased, focusing on narrow tasks and failing to assess the impact on in-domain samples. To address these issues, we introduce ComprehendEdit, a comprehensive benchmark comprising eight diverse tasks from multiple datasets. We propose two novel metrics: Knowledge Generalization Index (KGI) and Knowledge Preservation Index (KPI), which evaluate editing effects on in-domain samples without relying on AI-synthetic samples. Based on insights from our framework, we establish Hierarchical In-Context Editing (HICE), a baseline method employing a two-stage approach that balances performance across all metrics. This study provides a more comprehensive evaluation framework for multimodal knowledge editing, reveals unique challenges in this field, and offers a baseline method demonstrating improved performance. Our work opens new perspectives for future research and provides a foundation for developing more robust and effective editing techniques for MLLMs. The ComprehendEdit benchmark and implementation code are available at https://github.com/yaohui120/ComprehendEdit.

Anjia Cao, Xing Wei, Zhiheng Ma

Language-image pre-training faces significant challenges due to limited data in specific formats and the constrained capacities of text encoders. While prevailing methods attempt to address these issues through data augmentation and architecture modifications, they continue to struggle with processing long-form text inputs, and the inherent limitations of traditional CLIP text encoders lead to suboptimal downstream generalization. In this paper, we propose FLAME (Frozen Large lAnguage Models Enable data-efficient language-image pre-training) that leverages frozen large language models as text encoders, naturally processing long text inputs and demonstrating impressive multilingual generalization. FLAME comprises two key components: 1) a multifaceted prompt distillation technique for extracting diverse semantic representations from long captions, which better aligns with the multifaceted nature of images, and 2) a facet-decoupled attention mechanism, complemented by an offline embedding strategy, to ensure efficient computation. Extensive empirical evaluations demonstrate FLAME's superior performance. When trained on CC3M, FLAME surpasses the previous state-of-the-art by 4.9% in ImageNet top-1 accuracy. On YFCC15M, FLAME surpasses the WIT-400M-trained CLIP by 44.4\% in average image-to-text recall@1 across 36 languages, and by 34.6% in text-to-image recall@1 for long-context retrieval on Urban-1k. Code is available at https://github.com/MIV-XJTU/FLAME.

Hui Lin, Zhiheng Ma, Rongrong Ji et al.

This paper focuses on semi-supervised crowd counting, where only a small portion of the training data are labeled. We formulate the pixel-wise density value to regress as a probability distribution, instead of a single deterministic value. On this basis, we propose a semi-supervised crowd-counting model. Firstly, we design a pixel-wise distribution matching loss to measure the differences in the pixel-wise density distributions between the prediction and the ground truth; Secondly, we enhance the transformer decoder by using density tokens to specialize the forwards of decoders w.r.t. different density intervals; Thirdly, we design the interleaving consistency self-supervised learning mechanism to learn from unlabeled data efficiently. Extensive experiments on four datasets are performed to show that our method clearly outperforms the competitors by a large margin under various labeled ratio settings. Code will be released at https://github.com/LoraLinH/Semi-supervised-Counting-via-Pixel-by-pixel-Density-Distribution-Modelling.

Detao Bai, Shimin Yao, Weixuan Chen et al.

While Omni-modal Large Language Models have made strides in joint sensory processing, they fundamentally struggle with a cornerstone of human interaction: deciphering complex, multi-person conversational dynamics to accurately answer ``Who said what and when.'' Current models suffer from an ``illusion of competence'' -- they exploit visual biases in conventional benchmarks to bypass genuine cross-modal alignment, while relying on sparse, low-frame-rate visual sampling that destroys crucial high-frequency dynamics like lip movements. To shatter this illusion, we introduce Visual-Registered Speaker Diarization and Recognition (VR-SDR) and the HumanOmni-Speaker Benchmark. By strictly eliminating visual shortcuts, this rigorous paradigm demands true end-to-end spatio-temporal identity binding using only natural language queries. To overcome the underlying architectural perception gap, we propose HumanOmni-Speaker, powered by a Visual Delta Encoder. By sampling raw video at 25 fps and explicitly compressing inter-frame motion residuals into just 6 tokens per frame, it captures fine-grained visemes and speaker trajectories without triggering a catastrophic token explosion. Ultimately, HumanOmni-Speaker demonstrates strong multimodal synergy, natively enabling end-to-end lip-reading and high-precision spatial localization without intrusive cropping, and achieving superior performance across a wide spectrum of speaker-centric tasks.

Huoren Yang, Jianchao Zhao, Hu Yusong et al.

Vision-Language-Action (VLA) models have advanced rapidly with stronger backbones, broader pre-training, and larger demonstration datasets, yet their action heads remain largely homogeneous: most directly predict action commands in a fixed world coordinate frame. We propose \textbf{MCF-Proto}, a lightweight action head that equips VLA policies with a Motion-Centric Action Frame (MCF) and a prototype-based action parameterization. At each step, the policy predicts a rotation $R_t \in SO(3)$, composes actions in the transformed local frame from a set of prototypes, and maps them back to the world frame for end-to-end training, using only standard demonstrations without auxiliary supervision. This simple design induces stable emergent structure. Without explicit directional labels, the learned local frames develop a stable geometric structure whose axes are strongly compatible with demonstrated end-effector motion. Meanwhile, actions in the learned representation become substantially more compact, with variation captured by fewer dominant directions and more regularly organized by shared prototypes. These structural properties translate into improved robustness, especially under geometric perturbations. Our results suggest that adding lightweight geometric and compositional structure to the action head can materially improve how VLA policies organize and generalize robotic manipulation behavior. An anonymized code repository is provided in the supplementary material.

Junjin Xiao, Dongyang Li, Yandan Yang et al.

This paper tackles spatial perception and manipulation challenges in Vision-Language-Action (VLA) models. To address depth ambiguity from monocular input, we leverage a pre-trained multi-view diffusion model to synthesize latent novel views and propose a Geometry-Guided Gated Transformer (G3T) that aligns multi-view features under 3D geometric guidance while adaptively filtering occlusion noise. To improve action learning efficiency, we introduce Action Manifold Learning (AML), which directly predicts actions on the valid action manifold, bypassing inefficient regression of unstructured targets like noise or velocity. Experiments on LIBERO, RoboTwin 2.0, and real-robot tasks show our method achieves superior success rate and robustness over SOTA baselines. Project page: https://junjxiao.github.io/Multi-view-VLA.github.io/.

Jianchao Zhao, Huoren Yang, Hu Yusong et al.

Vision-Language-Action (VLA) models show strong potential for general-purpose robotic manipulation, yet their closed-loop reliability often degrades under local deployment conditions. Existing evaluations typically treat test episodes as independent zero-shot trials. However, real robots often operate repeatedly in the same or slowly changing environments, where successful executions provide environment-verified evidence of reliable behavior patterns. We study this persistent-deployment setting, asking whether a partially competent frozen VLA can improve its reliability by reusing its successful test-time experience. We propose an online success-memory guided test-time adaptation framework for generative VLAs. During deployment, the robot stores progress-calibrated successful observation-action segments in a long-term memory. At inference, it retrieves state-relevant action chunks, filters inconsistent candidates via trajectory-level consistency, and aggregates them into an elite action prior. To incorporate this prior into action generation, we introduce confidence-adaptive prior guidance, which injects the elite prior into an intermediate state of the flow-matching action sampler and adjusts the guidance strength based on retrieval confidence. This design allows the frozen VLA to exploit environment-specific successful experience while preserving observation-conditioned generative refinement. This retrieve-then-steer mechanism enables lightweight, non-parametric test-time adaptation without requiring parameter updates. Simulation and real-world experiments show improved task success and closed-loop stability, especially in long-horizon and multi-stage tasks.

Zuojin Tang, Haoyun Liu, Xinyuan Chang et al.

Vision-language-action (VLA) models remain constrained by the scarcity of action-labeled robot data, whereas action-free videos provide abundant evidence of how the physical world changes. Latent action models offer a promising way to extract such priors from videos, but reconstruction-trained latent codes are not necessarily suitable for policy generation: they may predict future observations while lacking the structure needed to be reused or generated coherently with robot actions. We introduce ALAM (Algebraic Latent Action Model), an Algebraically Consistent Latent Action Model that turns temporal relations in action-free video into structural supervision. Given frame triplets, ALAM learns latent transitions that are grounded by reconstruction while being regularized by composition and reversal consistency, encouraging a locally additive transition space. For downstream VLA learning, we freeze the pretrained encoder and use its latent transition sequences as auxiliary generative targets, co-generated with robot actions under a joint flow-matching objective. This couples structured latent transitions with flow-based policy generation, allowing the policy to exploit ALAM's locally consistent transition geometry without requiring latent-to-action decoding. Representation probes show that ALAM reduces additivity and reversibility errors by 25-85 times over unstructured latent-action baselines and improves long-horizon cumulative reconstruction. When transferred to VLA policies, ALAM raises the average success rate from 47.9% to 85.0% on MetaWorld MT50 and from 94.1% to 98.1% on LIBERO, with consistent gains on real-world manipulation tasks. Ablations further confirm that the strongest improvements arise from the synergy between algebraically structured latent transitions and joint flow matching.

Detao Bai, Zhiheng Ma, Xihan Wei et al.

The inherent synchronization between a speaker's lip movements, voice, and the underlying linguistic content offers a rich source of information for improving speech processing tasks, especially in challenging conditions where traditional audio-only systems falter. We introduce CoGenAV, a powerful and data-efficient model designed to learn versatile audio-visual representations applicable across a wide range of speech and audio-visual tasks. CoGenAV is trained by optimizing a dual objective derived from natural audio-visual synchrony, contrastive feature alignment and generative text prediction, using only 223 hours of labeled data from the LRS2 dataset. This contrastive-generative synchronization strategy effectively captures fundamental cross-modal correlations. We showcase the effectiveness and versatility of the learned CoGenAV representations on multiple benchmarks. When utilized for Audio-Visual Speech Recognition (AVSR) on LRS2, these representations contribute to achieving a state-of-the-art Word Error Rate (WER) of 1.27. They also enable strong performance in Visual Speech Recognition (VSR) with a WER of 20.5 on LRS2, and significantly improve performance in noisy environments by over 70%. Furthermore, CoGenAV representations benefit speech reconstruction tasks, boosting performance in Speech Enhancement and Separation, and achieve competitive results in audio-visual synchronization tasks like Active Speaker Detection (ASD). Our model will be open-sourced to facilitate further development and collaboration within both academia and industry.

Zhiheng Ma, Xing Wei, Xiaopeng Hong et al.

In crowd counting datasets, each person is annotated by a point, which is usually the center of the head. And the task is to estimate the total count in a crowd scene. Most of the state-of-the-art methods are based on density map estimation, which convert the sparse point annotations into a "ground truth" density map through a Gaussian kernel, and then use it as the learning target to train a density map estimator. However, such a "ground-truth" density map is imperfect due to occlusions, perspective effects, variations in object shapes, etc. On the contrary, we propose \emph{Bayesian loss}, a novel loss function which constructs a density contribution probability model from the point annotations. Instead of constraining the value at every pixel in the density map, the proposed training loss adopts a more reliable supervision on the count expectation at each annotated point. Without bells and whistles, the loss function makes substantial improvements over the baseline loss on all tested datasets. Moreover, our proposed loss function equipped with a standard backbone network, without using any external detectors or multi-scale architectures, plays favourably against the state of the arts. Our method outperforms previous best approaches by a large margin on the latest and largest UCF-QNRF dataset. The source code is available at \url{https://github.com/ZhihengCV/Baysian-Crowd-Counting}.

Haisen He, Xiangyu Zou, SongLin Dong et al.

Real-world image degradation is often unknown, spatially non-uniform, and compositional, requiring all-in-one restoration models to adapt a single set of weights to diverse local corruption patterns without test-time degradation labels. Existing methods typically modulate a shared backbone with global prompts or degradation descriptors, or route features through predefined expert pools. However, compact global conditioning can bottleneck localized degradation evidence, while static expert routing may produce homogeneous updates or rely on unstable sparse assignments. We propose \textbf{Continuous Expert Assembly} (CEA), a token-wise dynamic parameterization framework for all-in-one image restoration. CEA employs a lightweight \textbf{Cross-Attention Hyper-Adapter} to probe intermediate spatial features and synthesize instance-conditioned low-rank routing bases and residual directions. Each spatial token then assembles its own residual update via dense signed dot-product affinities over the generated rank-wise components, avoiding external prompts, static expert banks, and discrete Top- selection. The resulting assembly rule also admits a linear-attention perspective, making its dense token-wise routing behavior transparent. Experiments on AIO-3, AIO-5, and CDD-11 show that CEA improves average restoration quality over strong prompt-, descriptor-, and expert-based baselines, with the clearest gains on spatially varying and compositional degradations, while maintaining favorable parameter, FLOP, and runtime efficiency.

Shenxing Wei, Xing Wei, Zhiheng Ma et al.

Detecting anomaly patterns from images is a crucial artificial intelligence technique in industrial applications. Recent research in this domain has emphasized the necessity of a large volume of training data, overlooking the practical scenario where, post-deployment of the model, unlabeled data containing both normal and abnormal samples can be utilized to enhance the model's performance. Consequently, this paper focuses on addressing the challenging yet practical few-shot online anomaly detection and segmentation (FOADS) task. Under the FOADS framework, models are trained on a few-shot normal dataset, followed by inspection and improvement of their capabilities by leveraging unlabeled streaming data containing both normal and abnormal samples simultaneously. To tackle this issue, we propose modeling the feature distribution of normal images using a Neural Gas network, which offers the flexibility to adapt the topology structure to identify outliers in the data flow. In order to achieve improved performance with limited training samples, we employ multi-scale feature embedding extracted from a CNN pre-trained on ImageNet to obtain a robust representation. Furthermore, we introduce an algorithm that can incrementally update parameters without the need to store previous samples. Comprehensive experimental results demonstrate that our method can achieve substantial performance under the FOADS setting, while ensuring that the time complexity remains within an acceptable range on MVTec AD and BTAD datasets.

Zekai Li, Xinhao Zhong, Samir Khaki et al.

In recent years, dataset distillation has provided a reliable solution for data compression, where models trained on the resulting smaller synthetic datasets achieve performance comparable to those trained on the original datasets. To further improve the performance of synthetic datasets, various training pipelines and optimization objectives have been proposed, greatly advancing the field of dataset distillation. Recent decoupled dataset distillation methods introduce soft labels and stronger data augmentation during the post-evaluation phase and scale dataset distillation up to larger datasets (e.g., ImageNet-1K). However, this raises a question: Is accuracy still a reliable metric to fairly evaluate dataset distillation methods? Our empirical findings suggest that the performance improvements of these methods often stem from additional techniques rather than the inherent quality of the images themselves, with even randomly sampled images achieving superior results. Such misaligned evaluation settings severely hinder the development of DD. Therefore, we propose DD-Ranking, a unified evaluation framework, along with new general evaluation metrics to uncover the true performance improvements achieved by different methods. By refocusing on the actual information enhancement of distilled datasets, DD-Ranking provides a more comprehensive and fair evaluation standard for future research advancements.

Yandan Yang, Shuang Zeng, Tong Lin et al.

Building general-purpose embodied agents across diverse hardware remains a central challenge in robotics, often framed as the ''one-brain, many-forms'' paradigm. Progress is hindered by fragmented data, inconsistent representations, and misaligned training objectives. We present ABot-M0, a framework that builds a systematic data curation pipeline while jointly optimizing model architecture and training strategies, enabling end-to-end transformation of heterogeneous raw data into unified, efficient representations. From six public datasets, we clean, standardize, and balance samples to construct UniACT-dataset, a large-scale dataset with over 6 million trajectories and 9,500 hours of data, covering diverse robot morphologies and task scenarios. Unified pre-training improves knowledge transfer and generalization across platforms and tasks, supporting general-purpose embodied intelligence. To improve action prediction efficiency and stability, we propose the Action Manifold Hypothesis: effective robot actions lie not in the full high-dimensional space but on a low-dimensional, smooth manifold governed by physical laws and task constraints. Based on this, we introduce Action Manifold Learning (AML), which uses a DiT backbone to predict clean, continuous action sequences directly. This shifts learning from denoising to projection onto feasible manifolds, improving decoding speed and policy stability. ABot-M0 supports modular perception via a dual-stream mechanism that integrates VLM semantics with geometric priors and multi-view inputs from plug-and-play 3D modules such as VGGT and Qwen-Image-Edit, enhancing spatial understanding without modifying the backbone and mitigating standard VLM limitations in 3D reasoning. Experiments show components operate independently with additive benefits. We will release all code and pipelines for reproducibility and future research.

Yong Dai, Xiaopeng Hong, Yabin Wang et al.

Contemporary continual learning approaches typically select prompts from a pool, which function as supplementary inputs to a pre-trained model. However, this strategy is hindered by the inherent noise of its selection approach when handling increasing tasks. In response to these challenges, we reformulate the prompting approach for continual learning and propose the prompt customization (PC) method. PC mainly comprises a prompt generation module (PGM) and a prompt modulation module (PMM). In contrast to conventional methods that employ hard prompt selection, PGM assigns different coefficients to prompts from a fixed-sized pool of prompts and generates tailored prompts. Moreover, PMM further modulates the prompts by adaptively assigning weights according to the correlations between input data and corresponding prompts. We evaluate our method on four benchmark datasets for three diverse settings, including the class, domain, and task-agnostic incremental learning tasks. Experimental results demonstrate consistent improvement (by up to 16.2\%), yielded by the proposed method, over the state-of-the-art (SOTA) techniques.

Yabin Wang, Zhiwu Huang, Zhiheng Ma et al.

The emergence of text-to-image generative models has revolutionized the field of deepfakes, enabling the creation of realistic and convincing visual content directly from textual descriptions. However, this advancement presents considerably greater challenges in detecting the authenticity of such content. Existing deepfake detection datasets and methods often fall short in effectively capturing the extensive range of emerging deepfakes and offering satisfactory explanatory information for detection. To address the significant issue, this paper introduces a deepfake database (DFLIP-3K) for the development of convincing and explainable deepfake detection. It encompasses about 300K diverse deepfake samples from approximately 3K generative models, which boasts the largest number of deepfake models in the literature. Moreover, it collects around 190K linguistic footprints of these deepfakes. The two distinguished features enable DFLIP-3K to develop a benchmark that promotes progress in linguistic profiling of deepfakes, which includes three sub-tasks namely deepfake detection, model identification, and prompt prediction. The deepfake model and prompt are two essential components of each deepfake, and thus dissecting them linguistically allows for an invaluable exploration of trustworthy and interpretable evidence in deepfake detection, which we believe is the key for the next-generation deepfake detection. Furthermore, DFLIP-3K is envisioned as an open database that fosters transparency and encourages collaborative efforts to further enhance its growth. Our extensive experiments on the developed benchmark verify that our DFLIP-3K database is capable of serving as a standardized resource for evaluating and comparing linguistic-based deepfake detection, identification, and prompt prediction techniques.

Lingfeng He, De Cheng, Zhiheng Ma et al.

Continual Learning (CL) empowers AI models to continuously learn from sequential task streams. Recently, parameter-efficient fine-tuning (PEFT)-based CL methods have garnered increasing attention due to their superior performance. They typically allocate a unique sub-module for learning each task, with a task recognizer to select the appropriate sub-modules for testing images. However, due to the feature subspace misalignment from independently trained sub-modules, these methods tend to produce ambiguous decisions under misleading task-ids. To address this, we propose Cross-subspace Knowledge Alignment and Aggregation (CKAA), a novel framework that enhances model robustness against misleading task-ids through two key innovations: (1) Dual-level Knowledge Alignment (DKA): By aligning intra-class feature distributions across different subspaces and learning a robust global classifier through a feature simulation process, DKA enables the model to distinguish features from both correct and incorrect subspaces during training. (2) Task-Confidence-guided Mixture of Adapters (TC-MoA): A robust inference scheme that adaptively aggregates task-specific knowledge from relevant sub-modules based on task-confidence scores, avoiding overconfidence in misleading task-id predictions. Extensive experiments demonstrate that CKAA outperforms existing PEFT-based CL methods.

Songlin Dong, Jiangyang Li, Chenhao Ding et al.

Multi-label Class-Incremental Learning aims to continuously recognize novel categories in complex scenes where multiple objects co-occur. However, existing approaches often incur high computational costs due to full-parameter fine-tuning and substantial storage overhead from memory buffers, or they struggle to address feature confusion and domain discrepancies adequately. To overcome these limitations, we introduce P2L-CA, a parameter-efficient framework that integrates a Prompt-to-Label module with a Continuous Adapter module. The P2L module leverages class-specific prompts to disentangle multi-label representations while incorporating linguistic priors to enforce stable semantic-visual alignment. Meanwhile, the CA module employs lightweight adapters to mitigate domain gaps between pre-trained models and downstream tasks, thereby enhancing model plasticity. Extensive experiments across standard and challenging MLCIL settings on MS-COCO and PASCAL VOC show that P2L-CA not only achieves substantial improvements over state-of-the-art methods but also demonstrates strong generalization in CIL scenarios, all while requiring minimal trainable parameters and eliminating the need for memory buffers.

Ning Li, Kaitao Jiang, Zhiheng Ma et al.

Anomaly detection plays a key role in industrial manufacturing for product quality control. Traditional methods for anomaly detection are rule-based with limited generalization ability. Recent methods based on supervised deep learning are more powerful but require large-scale annotated datasets for training. In practice, abnormal products are rare thus it is very difficult to train a deep model in a fully supervised way. In this paper, we propose a novel unsupervised anomaly detection approach based on Self-organizing Map (SOM). Our method, Self-organizing Map for Anomaly Detection (SOMAD) maintains normal characteristics by using topological memory based on multi-scale features. SOMAD achieves state-of the-art performance on unsupervised anomaly detection and localization on the MVTec dataset.

Hui Lin, Xiaopeng Hong, Zhiheng Ma et al.

Traditional crowd counting approaches usually use Gaussian assumption to generate pseudo density ground truth, which suffers from problems like inaccurate estimation of the Gaussian kernel sizes. In this paper, we propose a new measure-based counting approach to regress the predicted density maps to the scattered point-annotated ground truth directly. First, crowd counting is formulated as a measure matching problem. Second, we derive a semi-balanced form of Sinkhorn divergence, based on which a Sinkhorn counting loss is designed for measure matching. Third, we propose a self-supervised mechanism by devising a Sinkhorn scale consistency loss to resist scale changes. Finally, an efficient optimization method is provided to minimize the overall loss function. Extensive experiments on four challenging crowd counting datasets namely ShanghaiTech, UCF-QNRF, JHU++, and NWPU have validated the proposed method.