Zhe Zhang

Alisson Azzolini, Junjie Bai, Hannah Brandon et al. · nvidia

Physical AI systems need to perceive, understand, and perform complex actions in the physical world. In this paper, we present the Cosmos-Reason1 models that can understand the physical world and generate appropriate embodied decisions (e.g., next step action) in natural language through long chain-of-thought reasoning processes. We begin by defining key capabilities for Physical AI reasoning, with a focus on physical common sense and embodied reasoning. To represent physical common sense, we use a hierarchical ontology that captures fundamental knowledge about space, time, and physics. For embodied reasoning, we rely on a two-dimensional ontology that generalizes across different physical embodiments. Building on these capabilities, we develop two multimodal large language models, Cosmos-Reason1-7B and Cosmos-Reason1-56B. We curate data and train our models in two stages: Physical AI supervised fine-tuning (SFT) and Physical AI reinforcement learning (RL). To evaluate our models, we build comprehensive benchmarks for physical common sense and embodied reasoning according to our ontologies. Evaluation results show that Physical AI SFT and RL bring significant improvements. To facilitate the development of Physical AI, we make our code and pre-trained models available under the NVIDIA Open Model License at https://github.com/nvidia-cosmos/cosmos-reason1.

Benjamin Le, Xueying Lu, Nick Stern et al.

Evaluating relevance in large-scale search systems is fundamentally constrained by the governance gap between nuanced, resource-constrained human oversight and the high-throughput requirements of production systems. While traditional approaches rely on engagement proxies or sparse manual review, these methods often fail to capture the full scope of high-impact relevance failures. We present \textbf{SAGE} (Scalable AI Governance \& Evaluation), a framework that operationalizes high-quality human product judgment as a scalable evaluation signal. At the core of SAGE is a bidirectional calibration loop where natural-language \emph{Policy}, curated \emph{Precedent}, and an \emph{LLM Surrogate Judge} co-evolve. SAGE systematically resolves semantic ambiguities and misalignments, transforming subjective relevance judgment into an executable, multi-dimensional rubric with near human-level agreement. To bridge the gap between frontier model reasoning and industrial-scale inference, we apply teacher-student distillation to transfer high-fidelity judgments into compact student surrogates at \textbf{92$\times$} lower cost. Deployed within LinkedIn Search ecosystems, SAGE guided model iteration through simulation-driven development, distilling policy-aligned models for online serving and enabling rapid offline evaluation. In production, it powered policy oversight that measured ramped model variants and detected regressions invisible to engagement metrics. Collectively, these drove a \textbf{0.25\%} lift in LinkedIn daily active users.

Xingyu Qiu, Yuqian Fu, Jiawei Geng et al.

Cross-domain few-shot object detection (CD-FSOD) remains a challenging problem for existing object detectors and few-shot learning approaches, particularly when generalizing across distinct domains. As part of NTIRE 2026, we hosted the second CD-FSOD Challenge to systematically evaluate and promote progress in detecting objects in unseen target domains under limited annotation conditions. The challenge received strong community interest, with 128 registered participants and a total of 696 submissions. Among them, 31 teams actively participated, and 19 teams submitted valid final results. Participants explored a wide range of strategies, introducing innovative methods that push the performance frontier under both open-source and closed-source tracks. This report presents a detailed overview of the NTIRE 2026 CD-FSOD Challenge, including a summary of the submitted approaches and an analysis of the final results across all participating teams. Challenge Codes: https://github.com/ohMargin/NTIRE2026_CDFSOD.

Zitao Song, Cedar Site Bai, Zhe Zhang et al.

Training instabilities such as loss spikes are frequently the result of stochastic gradient noise. Because of rare expressions in language training data, and multiple layer composition, the noise impact is heavy-tailed and survives mini-batch averaging. Existing remedies trade off structure against cost: vector-norm clipping ignores the matrix structure of weight updates, while spectral normalization (e.g., Muon (Jordan et al., 2024)) respects it at additional cost. We show that this trade-off can be balanced. Real gradient noise appears to be similar to entry-wise heavy-tailed contamination, and a first-order perturbation analysis reveals a localization property of such noise, under which a simple entry-wise method achieves spectral control. Exploiting this, we derive a tractable surrogate for the Bayes-optimal entry-wise estimator under a Gaussian signal prior. We establish $O(ε^{-4})$ convergence guarantee under Cauchy-contaminated noise. Empirically, we find that smooth shrinkage improves Adam on NanoGPT pretraining, saving ${\sim}7\%$ of training tokens. We further find that applying the entry-wise clipping before spectral normalization yields a ${\sim}2\%$ token saving on top of Muon.

Zhe Zhang, Gaochang Wu, Jing Zhang et al.

Unsupervised Domain Adaptive Semantic Segmentation (UDA-SS) aims to transfer the supervision from a labeled source domain to an unlabeled target domain. The majority of existing UDA-SS works typically consider images whilst recent attempts have extended further to tackle videos by modeling the temporal dimension. Although the two lines of research share the major challenges -- overcoming the underlying domain distribution shift, their studies are largely independent, resulting in fragmented insights, a lack of holistic understanding, and missed opportunities for cross-pollination of ideas. This fragmentation prevents the unification of methods, leading to redundant efforts and suboptimal knowledge transfer across image and video domains. Under this observation, we advocate unifying the study of UDA-SS across video and image scenarios, enabling a more comprehensive understanding, synergistic advancements, and efficient knowledge sharing. To that end, we explore the unified UDA-SS from a general data augmentation perspective, serving as a unifying conceptual framework, enabling improved generalization, and potential for cross-pollination of ideas, ultimately contributing to the overall progress and practical impact of this field of research. Specifically, we propose a Quad-directional Mixup (QuadMix) method, characterized by tackling distinct point attributes and feature inconsistencies through four-directional paths for intra- and inter-domain mixing in a feature space. To deal with temporal shifts with videos, we incorporate optical flow-guided feature aggregation across spatial and temporal dimensions for fine-grained domain alignment. Extensive experiments show that our method outperforms the state-of-the-art works by large margins on four challenging UDA-SS benchmarks. Our source code and models will be released at https://github.com/ZHE-SAPI/UDASS.

Kan Xie, Zhe Zhang, Bo Li et al.

With the gradual popularization of self-driving, it is becoming increasingly important for vehicles to smartly make the right driving decisions and autonomously obey traffic rules by correctly recognizing traffic signs. However, for machine learning-based traffic sign recognition on the Internet of Vehicles (IoV), a large amount of traffic sign data from distributed vehicles is needed to be gathered in a centralized server for model training, which brings serious privacy leakage risk because of traffic sign data containing lots of location privacy information. To address this issue, we first exploit privacy-preserving federated learning to perform collaborative training for accurate recognition models without sharing raw traffic sign data. Nevertheless, due to the limited computing and energy resources of most devices, it is hard for vehicles to continuously undertake complex artificial intelligence tasks. Therefore, we introduce powerful Spike Neural Networks (SNNs) into traffic sign recognition for energy-efficient and fast model training, which is the next generation of neural networks and is practical and well-fitted to IoV scenarios. Furthermore, we design a novel encoding scheme for SNNs based on neuron receptive fields to extract information from the pixel and spatial dimensions of traffic signs to achieve high-accuracy training. Numerical results indicate that the proposed federated SNN outperforms traditional federated convolutional neural networks in terms of accuracy, noise immunity, and energy efficiency as well.

Chaohui Yu, Qiang Zhou, Jingliang Li et al.

Text-to-3D generation has recently garnered significant attention, fueled by 2D diffusion models trained on billions of image-text pairs. Existing methods primarily rely on score distillation to leverage the 2D diffusion priors to supervise the generation of 3D models, e.g., NeRF. However, score distillation is prone to suffer the view inconsistency problem, and implicit NeRF modeling can also lead to an arbitrary shape, thus leading to less realistic and uncontrollable 3D generation. In this work, we propose a flexible framework of Points-to-3D to bridge the gap between sparse yet freely available 3D points and realistic shape-controllable 3D generation by distilling the knowledge from both 2D and 3D diffusion models. The core idea of Points-to-3D is to introduce controllable sparse 3D points to guide the text-to-3D generation. Specifically, we use the sparse point cloud generated from the 3D diffusion model, Point-E, as the geometric prior, conditioned on a single reference image. To better utilize the sparse 3D points, we propose an efficient point cloud guidance loss to adaptively drive the NeRF's geometry to align with the shape of the sparse 3D points. In addition to controlling the geometry, we propose to optimize the NeRF for a more view-consistent appearance. To be specific, we perform score distillation to the publicly available 2D image diffusion model ControlNet, conditioned on text as well as depth map of the learned compact geometry. Qualitative and quantitative comparisons demonstrate that Points-to-3D improves view consistency and achieves good shape controllability for text-to-3D generation. Points-to-3D provides users with a new way to improve and control text-to-3D generation.

Muhan Wang, Zhe Zhang, Xiaolan Qiu et al.

Tomographic SAR technique has attracted remarkable interest for its ability of three-dimensional resolving along the elevation direction via a stack of SAR images collected from different cross-track angles. The emerged compressed sensing (CS)-based algorithms have been introduced into TomoSAR considering its super-resolution ability with limited samples. However, the conventional CS-based methods suffer from several drawbacks, including weak noise resistance, high computational complexity, and complex parameter fine-tuning. Aiming at efficient TomoSAR imaging, this paper proposes a novel efficient sparse unfolding network based on the analytic learned iterative shrinkage thresholding algorithm (ALISTA) architecture with adaptive threshold, named Adaptive Threshold ALISTA-based Sparse Imaging Network (ATASI-Net). The weight matrix in each layer of ATASI-Net is pre-computed as the solution of an off-line optimization problem, leaving only two scalar parameters to be learned from data, which significantly simplifies the training stage. In addition, adaptive threshold is introduced for each azimuth-range pixel, enabling the threshold shrinkage to be not only layer-varied but also element-wise. Moreover, the final learned thresholds can be visualized and combined with the SAR image semantics for mutual feedback. Finally, extensive experiments on simulated and real data are carried out to demonstrate the effectiveness and efficiency of the proposed method.

Zitao Song, Cedar Site Bai, Zhe Zhang et al.

Adaptive methods like Adam have become the $\textit{de facto}$ standard for large-scale vector and Euclidean optimization due to their coordinate-wise adaptation with a second-order nature. More recently, matrix-based spectral optimizers like Muon (Jordan et al., 2024b) show the power of treating weight matrices as matrices rather than long vectors. Linking these is hard because many natural generalizations are not feasible to implement, and we also cannot simply move the Adam adaptation to the matrix spectrum. To address this, we reformulate the AdaGrad update and decompose it into a variance adaptation term and a scale-invariant term. This decoupling produces $\textbf{DeVA}$ ($\textbf{De}$coupled $\textbf{V}$ariance $\textbf{A}$daptation), a framework that bridges between vector-based variance adaptation and matrix spectral optimization, enabling a seamless transition from Adam to adaptive spectral descent. Extensive experiments across language modeling and image classification demonstrate that DeVA consistently outperforms state-of-the-art methods such as Muon and SOAP (Vyas et al., 2024), reducing token usage by around 6.6\%. Theoretically, we show that the variance adaptation term effectively improves the blockwise smoothness, facilitating faster convergence. Our implementation is available at https://github.com/Tsedao/Decoupled-Variance-Adaptation

Jiacheng Deng, Chuxin Wang, Jiahao Lu et al.

Unsupervised point cloud shape correspondence aims to obtain dense point-to-point correspondences between point clouds without manually annotated pairs. However, humans and some animals have bilateral symmetry and various orientations, which lead to severe mispredictions of symmetrical parts. Besides, point cloud noise disrupts consistent representations for point cloud and thus degrades the shape correspondence accuracy. To address the above issues, we propose a Self-Ensembling ORientation-aware Network termed SE-ORNet. The key of our approach is to exploit an orientation estimation module with a domain adaptive discriminator to align the orientations of point cloud pairs, which significantly alleviates the mispredictions of symmetrical parts. Additionally, we design a selfensembling framework for unsupervised point cloud shape correspondence. In this framework, the disturbances of point cloud noise are overcome by perturbing the inputs of the student and teacher networks with different data augmentations and constraining the consistency of predictions. Extensive experiments on both human and animal datasets show that our SE-ORNet can surpass state-of-the-art unsupervised point cloud shape correspondence methods.

Vincent Jeanselme, Maria De-Arteaga, Zhe Zhang et al.

Machine learning risks reinforcing biases present in data and, as we argue in this work, in what is absent from data. In healthcare, societal and decision biases shape patterns in missing data, yet the algorithmic fairness implications of group-specific missingness are poorly understood. The way we address missingness in healthcare can have detrimental impacts on downstream algorithmic fairness. Our work questions current recommendations and practices aimed at handling missing data with a focus on their effect on algorithmic fairness, and offers a path forward. Specifically, we consider the theoretical underpinnings of existing recommendations as well as their empirical predictive performance and corresponding algorithmic fairness measured through subgroup performances. Our results show that current practices for handling missingness lack principled foundations, are disconnected from the realities of missingness mechanisms in healthcare, and can be counterproductive. For example, we show that favouring group-specific imputation strategy can be misguided and exacerbate prediction disparities. We then build on our findings to propose a framework for empirically guiding imputation choices, and an accompanying reporting framework. Our work constitutes an important contribution to recent efforts by regulators and practitioners to grapple with the realities of real-world data, and to foster the responsible and transparent deployment of machine learning systems. We demonstrate the practical utility of the proposed framework through experimentation on widely used datasets, where we show how the proposed framework can guide the selection of imputation strategies, allowing us to choose among strategies that yield equal overall predictive performance but present different algorithmic fairness properties.

Zhe Zhang, Xingrong Diao, Haoxiang Liang et al.

We propose an optimization-based framework for robust contact-rich manipulation. Recent contact-implicit methods enable online hybrid planning across contact modes, allowing closed-loop manipulation for a given target state and contact location sequence of the robot and object. However, most existing approaches lack the ability to autonomously reason and generate diverse contact location sequences and manipulation trajectories, i.e., active contact location selection, which limits their applicability to relatively simple tasks. Active contact location selection is challenging due to complementarity in contact dynamics and the sparse gradients, making the design of a unified framework for contact selection and planning difficult. To address these challenges, we introduce Simultaneous Contact Selection and Planning (SCSP), a cascaded optimization framework comprising Contact Selection Optimization (CSO) and Contact Planning Optimization (CPO). CSO leverages a surrogate contact model and discrete-continuous optimization to efficiently resolve the nonsmoothness and coupling in contact selection, enabling online global searching of optimal contact locations. CPO performs prior-guided contact planning by evaluating the reference contact locations produced by CSO and generating corresponding manipulation trajectories in real time for redundant manipulators. Extensive simulations and real-world experiments demonstrate that SCSP produces diverse manipulation behaviors and robust control under inaccurate dynamics and perceptual noise. We further validate the generalization of the framework on challenging manipulation tasks. Project website: \href{https://sites.google.com/view/scsp-robot}{https://sites.google.com/view/scsp-robot}.

Zhe Zhang, Yi Yu, Atsuhiro Takasu

Lyrics-to-melody generation is an interesting and challenging topic in AI music research field. Due to the difficulty of learning the correlations between lyrics and melody, previous methods suffer from low generation quality and lack of controllability. Controllability of generative models enables human interaction with models to generate desired contents, which is especially important in music generation tasks towards human-centered AI that can facilitate musicians in creative activities. To address these issues, we propose a controllable lyrics-to-melody generation network, ConL2M, which is able to generate realistic melodies from lyrics in user-desired musical style. Our work contains three main novelties: 1) To model the dependencies of music attributes cross multiple sequences, inter-branch memory fusion (Memofu) is proposed to enable information flow between multi-branch stacked LSTM architecture; 2) Reference style embedding (RSE) is proposed to improve the quality of generation as well as control the musical style of generated melodies; 3) Sequence-level statistical loss (SeqLoss) is proposed to help the model learn sequence-level features of melodies given lyrics. Verified by evaluation metrics for music quality and controllability, initial study of controllable lyrics-to-melody generation shows better generation quality and the feasibility of interacting with users to generate the melodies in desired musical styles when given lyrics.

Lei Kou, Chuang Liu, Guowei Cai et al.

A fault diagnosis method for power electronics converters based on deep feedforward network and wavelet compression is proposed in this paper. The transient historical data after wavelet compression are used to realize the training of fault diagnosis classifier. Firstly, the correlation analysis of the voltage or current data running in various fault states is performed to remove the redundant features and the sampling point. Secondly, the wavelet transform is used to remove the redundant data of the features, and then the training sample data is greatly compressed. The deep feedforward network is trained by the low frequency component of the features, while the training speed is greatly accelerated. The average accuracy of fault diagnosis classifier can reach over 97%. Finally, the fault diagnosis classifier is tested, and final diagnosis result is determined by multiple-groups transient data, by which the reliability of diagnosis results is improved. The experimental result proves that the classifier has strong generalization ability and can accurately locate the open-circuit faults in IGBTs.

Ryan Humble, Zhe Zhang, Finn O'Shea et al.

Anomaly detection is an important task for complex systems (e.g., industrial facilities, manufacturing, large-scale science experiments), where failures in a sub-system can lead to low yield, faulty products, or even damage to components. While complex systems often have a wealth of data, labeled anomalies are typically rare (or even nonexistent) and expensive to acquire. Unsupervised approaches are therefore common and typically search for anomalies either by distance or density of examples in the input feature space (or some associated low-dimensional representation). This paper presents a novel approach called CoAD, which is specifically designed for multi-modal tasks and identifies anomalies based on \textit{coincident} behavior across two different slices of the feature space. We define an \textit{unsupervised} metric, $\hat{F}_β$, out of analogy to the supervised classification $F_β$ statistic. CoAD uses $\hat{F}_β$ to train an anomaly detection algorithm on \textit{unlabeled data}, based on the expectation that anomalous behavior in one feature slice is coincident with anomalous behavior in the other. The method is illustrated using a synthetic outlier data set and a MNIST-based image data set, and is compared to prior state-of-the-art on two real-world tasks: a metal milling data set and a data set from a particle accelerator.

Chuang Liu, Lei Kou, Guowei Cai et al.

Power electronics converters have been widely used in aerospace system, DC transmission, distributed energy, smart grid and so forth, and the reliability of power electronics converters has been a hotspot in academia and industry. It is of great significance to carry out power electronics converters open-circuit faults monitoring and intelligent fault diagnosis to avoid secondary faults, reduce time and cost of operation and maintenance, and improve the reliability of power electronics system. Firstly, the faults features of power electronic converters are analyzed and summarized. Secondly, some AI-based fault diagnosis methods and application examples in power electronics converters are reviewed, and a fault diagnosis method based on the combination of random forests and transient fault features is proposed for three-phase power electronics converters. Finally, the future research challenges and directions of AI-based fault diagnosis methods are pointed out.

Mingxiu Cai, Zhe Zhang, Gaochang Wu et al.

Unsupervised Anomaly Detection (UAD) aims to identify abnormal regions by establishing correspondences between test images and normal templates. Existing methods primarily rely on image reconstruction or template retrieval but face a fundamental challenge: matching between test images and normal templates inevitably introduces noise due to intra-class variations, imperfect correspondences, and limited templates. Observing that Retrieval-Augmented Generation (RAG) leverages retrieved samples directly in the generation process, we reinterpret UAD through this lens and introduce \textbf{RAID}, a retrieval-augmented UAD framework designed for noise-resilient anomaly detection and localization. Unlike standard RAG that enriches context or knowledge, we focus on using retrieved normal samples to guide noise suppression in anomaly map generation. RAID retrieves class-, semantic-, and instance-level representations from a hierarchical vector database, forming a coarse-to-fine pipeline. A matching cost volume correlates the input with retrieved exemplars, followed by a guided Mixture-of-Experts (MoE) network that leverages the retrieved samples to adaptively suppress matching noise and produce fine-grained anomaly maps. RAID achieves state-of-the-art performance across full-shot, few-shot, and multi-dataset settings on MVTec, VisA, MPDD, and BTAD benchmarks. \href{https://github.com/Mingxiu-Cai/RAID}{https://github.com/Mingxiu-Cai/RAID}.

Guanghui Lan, Zhe Zhang

This paper studies the communication complexity of convex risk-averse optimization over a network. The problem generalizes the well-studied risk-neutral finite-sum distributed optimization problem and its importance stems from the need to handle risk in an uncertain environment. For algorithms in the literature, there exists a gap in communication complexities for solving risk-averse and risk-neutral problems. We propose two distributed algorithms, namely the distributed risk averse optimization (DRAO) method and the distributed risk averse optimization with sliding (DRAO-S) method, to close the gap. Specifically, the DRAO method achieves the optimal communication complexity by assuming a certain saddle point subproblem can be easily solved in the server node. The DRAO-S method removes the strong assumption by introducing a novel saddle point sliding subroutine which only requires the projection over the ambiguity set $P$. We observe that the number of $P$-projections performed by DRAO-S is optimal. Moreover, we develop matching lower complexity bounds to show the communication complexities of both DRAO and DRAO-S to be improvable. Numerical experiments are conducted to demonstrate the encouraging empirical performance of the DRAO-S method.

Haifeng Wang, Hua Wu, Tian Wu et al.

In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are trained from scratch under a unified next-group-of-tokens prediction objective, based on an ultra-sparse mixture-of-experts (MoE) architecture with modality-agnostic expert routing. To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm. Within a single pre-training run, the model learns a family of sub-models with varying depths, expert capacities, and routing sparsity, enabling flexible trade-offs among performance, model size, and inference latency in memory- or time-constrained scenarios. Moreover, we systematically address the challenges of scaling reinforcement learning to unified foundation models, thereby guaranteeing efficient and stable post-training under ultra-sparse MoE architectures and diverse multimodal settings. Extensive experiments demonstrate that ERNIE 5.0 achieves strong and balanced performance across multiple modalities. To the best of our knowledge, among publicly disclosed models, ERNIE 5.0 represents the first production-scale realization of a trillion-parameter unified autoregressive model that supports both multimodal understanding and generation. To facilitate further research, we present detailed visualizations of modality-agnostic expert routing in the unified model, alongside comprehensive empirical analysis of elastic training, aiming to offer profound insights to the community.

Yigitcan Özer, Wanying Ge, Zhe Zhang et al.

Audio watermarking embeds auxiliary information into speech while maintaining speaker identity, linguistic content, and perceptual quality. Although recent advances in neural and digital signal processing-based watermarking methods have improved imperceptibility and embedding capacity, robustness is still primarily assessed against conventional distortions such as compression, additive noise, and resampling. However, the rise of deep learning-based attacks introduces novel and significant threats to watermark security. In this work, we investigate self voice conversion as a universal, content-preserving attack against audio watermarking systems. Self voice conversion remaps a speaker's voice to the same identity while altering acoustic characteristics through a voice conversion model. We demonstrate that this attack severely degrades the reliability of state-of-the-art watermarking approaches and highlight its implications for the security of modern audio watermarking techniques.

Jinghui Xiao, Qun Liu, Xin Jiang et al.

Pinyin to Character conversion (P2C) task is the key task of Input Method Engine (IME) in commercial input software for Asian languages, such as Chinese, Japanese, Thai language and so on. It's usually treated as sequence labelling task and resolved by language model, i.e. n-gram or RNN. However, the low capacity of the n-gram or RNN limits its performance. This paper introduces a new solution named PERT which stands for bidirectional Pinyin Encoder Representations from Transformers. It achieves significant improvement of performance over baselines. Furthermore, we combine PERT with n-gram under a Markov framework, and improve performance further. Lastly, the external lexicon is incorporated into PERT so as to resolve the OOD issue of IME.

Zhe Zhang, Martijn Goorden, Michel Reniers

Existing literature on timed opacity uses specific definitions for restricted subclasses of timed automata or limited observation models. This lack of a unified definition makes it difficult to establish formal relationships and compare the expressiveness of different opacity variants. This paper establishes a unified framework for timed opacity by introducing a universal observation model for timed automata. First, we introduce an observation model with full observation of time delay and partial observation of locations, clocks, and events. Second, based on this model, we define the notion of evolution-based timed opacity. Third, we mathematically prove that evolution-based timed opacity strictly implies language-based timed opacity and establish a formal equivalence with execution-time opacity under constrained observations. This framework establishes a unified semantic hierarchy for characterizing the landscape of timed opacity.

Zhe Zhang, Peijia Xie, Yuhan Pang et al.

Precise and flexible cart-pushing is a challenging task for mobile robots. The motion constraints during cart-pushing and the robot's redundancy lead to complex motion planning problems, while variable payloads and disturbances present complicated dynamics. In this work, we propose a novel planning and control framework for flexible whole-body coordination and robust adaptive control. Our motion planning method employs a local coordinate representation and a novel kinematic model to solve a nonlinear optimization problem, thereby enhancing motion maneuverability by generating feasible and flexible push poses. Furthermore, we present a disturbance rejection control method to resist disturbances and reduce control errors for the complex control problem without requiring an accurate dynamic model. We validate our method through extensive experiments in simulation and real-world settings, demonstrating its superiority over existing approaches. To the best of our knowledge, this is the first work to systematically evaluate the flexibility and robustness of cart-pushing methods in experiments. The video supplement is available at https://sites.google.com/view/mpac-pushing/.

Gurunath Reddy M, Zhe Zhang, Yi Yu et al.

We propose a deep attention-based alignment network, which aims to automatically predict lyrics and melody with given incomplete lyrics as input in a way similar to the music creation of humans. Most importantly, a deep neural lyrics-to-melody net is trained in an encoder-decoder way to predict possible pairs of lyrics-melody when given incomplete lyrics (few keywords). The attention mechanism is exploited to align the predicted lyrics with the melody during the lyrics-to-melody generation. The qualitative and quantitative evaluation metrics reveal that the proposed method is indeed capable of generating proper lyrics and corresponding melody for composing new songs given a piece of incomplete seed lyrics.

Shuoguang Yang, Wei You, Zhe Zhang et al.

Stochastic compositional optimization (SCO) has attracted considerable attention because of its broad applicability to important real-world problems. However, existing works on SCO assume that the projection within a solution update is simple, which fails to hold for problem instances where the constraints are in the form of expectations, such as empirical conditional value-at-risk constraints. We study a novel model that incorporates single-level expected value and two-level compositional constraints into the current SCO framework. Our model can be applied widely to data-driven optimization and risk management, including risk-averse optimization and high-moment portfolio selection, and can handle multiple constraints. We further propose a class of primal-dual algorithms that generates sequences converging to the optimal solution at the rate of $\cO(\frac{1}{\sqrt{N}})$under both single-level expected value and two-level compositional constraints, where $N$ is the iteration counter, establishing the benchmarks in expected value constrained SCO.

Chuyi Chen, Zhe Zhang, Yanchao Zhao

Federated learning (FL) has been widely adopted across various applications, such as healthcare, finance, and smart cities. However, as experimental scenarios become more complex, existing FL frameworks and benchmarks have struggled to keep pace. This paper introduces FedModule, a flexible and extensible FL experimental framework that has been open-sourced to support diverse FL paradigms and provide comprehensive benchmarks for complex experimental scenarios. FedModule adheres to the "one code, all scenarios" principle and employs a modular design that breaks the FL process into individual components, allowing for the seamless integration of different FL paradigms. The framework supports synchronous, asynchronous, and personalized federated learning, with over 20 implemented algorithms. Experiments conducted on public datasets demonstrate the flexibility and extensibility of FedModule. The framework offers multiple execution modes-including linear, threaded, process-based, and distributed-enabling users to tailor their setups to various experimental needs. Additionally, FedModule provides extensive logging and testing capabilities, which facilitate detailed performance analysis of FL algorithms. Comparative evaluations against existing FL toolkits, such as TensorFlow Federated, PySyft, Flower, and FLGo, highlight FedModule's superior scalability, flexibility, and comprehensive benchmark support. By addressing the limitations of current FL frameworks, FedModule marks a significant advancement in FL experimentation, providing researchers and practitioners with a robust tool for developing and evaluating FL algorithms across a wide range of scenarios.

Houming Chen, Zhe Zhang, H. V. Jagadish

Column matching is a central task in reconciling schemas for data integration. Column names and descriptions are valuable for this task. LLMs can leverage such natural-language schema metadata. However, in many datasets, correct matching requires additional evidence beyond the column itself. Because it is impractical to provide an LLM with the entire schema metadata needed to capture this evidence, the core challenge becomes to select and organize the most useful contextual information. We present ConStruM, a structure-guided framework for budgeted evidence packing in schema matching. ConStruM constructs a lightweight, reusable structure in which, at query time, it assembles a small context pack emphasizing the most discriminative evidence. ConStruM is designed as an add-on: given a shortlist of candidate targets produced by an upstream matcher, it augments the matcher's final LLM prompt with structured, query-specific evidence so that the final selection is better grounded. For this purpose, we develop a context tree for budgeted multi-level context retrieval and a global similarity hypergraph that surfaces groups of highly similar columns (on both the source and target sides), summarized via group-aware differentiation cues computed online or precomputed offline. Experiments on real datasets show that ConStruM improves matching by providing and organizing the right contextual evidence.

Zhe Zhang, Yuanning Feng, Yuxuan Song et al.

AlphaFold3 introduces a diffusion-based architecture that elevates protein structure prediction to all-atom resolution with improved accuracy. This state-of-the-art performance has established AlphaFold3 as a foundation model for diverse generation and design tasks. However, its iterative design substantially increases inference time, limiting practical deployment in downstream settings such as virtual screening and protein design. We propose DCFold, a single-step generative model that attains AlphaFold3-level accuracy. Our Dual Consistency training framework, which incorporates a novel Temporal Geodesic Matching (TGM) scheduler, enables DCFold to achieve a 15x acceleration in inference while maintaining predictive fidelity. We validate its effectiveness across both structure prediction and binder design benchmarks.

Zhe Zhang, Yukun Zou, Junjie Lai et al.

Deep Q-learning Network (DQN) is a successful way which combines reinforcement learning with deep neural networks and leads to a widespread application of reinforcement learning. One challenging problem when applying DQN or other reinforcement learning algorithms to real world problem is data collection. Therefore, how to improve data efficiency is one of the most important problems in the research of reinforcement learning. In this paper, we propose a framework which uses the Max-Mean loss in Deep Q-Network (M$^2$DQN). Instead of sampling one batch of experiences in the training step, we sample several batches from the experience replay and update the parameters such that the maximum TD-error of these batches is minimized. The proposed method can be combined with most of existing techniques of DQN algorithm by replacing the loss function. We verify the effectiveness of this framework with one of the most widely used techniques, Double DQN (DDQN), in several gym games. The results show that our method leads to a substantial improvement in both the learning speed and performance.

Zhe Zhang, Jing Li, Wanli Xue et al.

Assuming that neither source data nor the source model is accessible, black box domain adaptation represents a highly practical yet extremely challenging setting, as transferable information is restricted to the predictions of the black box source model, which can only be queried using target samples. Existing approaches attempt to extract transferable knowledge through pseudo label refinement or by leveraging external vision language models (ViLs), but they often suffer from noisy supervision or insufficient utilization of the semantic priors provided by ViLs, which ultimately hinder adaptation performance. To overcome these limitations, we propose a dual teacher distillation with subnetwork rectification (DDSR) model that jointly exploits the specific knowledge embedded in black box source models and the general semantic information of a ViL. DDSR adaptively integrates their complementary predictions to generate reliable pseudo labels for the target domain and introduces a subnetwork driven regularization strategy to mitigate overfitting caused by noisy supervision. Furthermore, the refined target predictions iteratively enhance both the pseudo labels and ViL prompts, enabling more accurate and semantically consistent adaptation. Finally, the target model is further optimized through self training with classwise prototypes. Extensive experiments on multiple benchmark datasets validate the effectiveness of our approach, demonstrating consistent improvements over state of the art methods, including those using source data or models.

Erle Zhu, Yadi Liu, Zhe Zhang et al.

Pre-trained on extensive text and image corpora, current Multi-Modal Large Language Models (MLLM) have shown strong capabilities in general visual reasoning tasks. However, their performance is still lacking in physical domains that require understanding diagrams with complex physical structures and quantitative analysis based on multi-modal information. To address this, we develop a new framework, named Multi-Modal Scientific Reasoning with Physics Perception and Simulation (MAPS) based on an MLLM. MAPS decomposes expert-level multi-modal reasoning task into physical diagram understanding via a Physical Perception Model (PPM) and reasoning with physical knowledge via a simulator. The PPM module is obtained by fine-tuning a visual language model using carefully designed synthetic data with paired physical diagrams and corresponding simulation language descriptions. At the inference stage, MAPS integrates the simulation language description of the input diagram provided by PPM and results obtained through a Chain-of-Simulation process with MLLM to derive the underlying rationale and the final answer. Validated using our collected college-level circuit analysis problems, MAPS significantly improves reasoning accuracy of MLLM and outperforms all existing models. The results confirm MAPS offers a promising direction for enhancing multi-modal scientific reasoning ability of MLLMs. We will release our code, model and dataset used for our experiments upon publishing of this paper.

Yaozhong Gan, Zhe Zhang, Xiaoyang Tan

Advantage learning (AL) aims to improve the robustness of value-based reinforcement learning against estimation errors with action-gap-based regularization. Unfortunately, the method tends to be unstable in the case of function approximation. In this paper, we propose a simple variant of AL, named smoothing advantage learning (SAL), to alleviate this problem. The key to our method is to replace the original Bellman Optimal operator in AL with a smooth one so as to obtain more reliable estimation of the temporal difference target. We give a detailed account of the resulting action gap and the performance bound for approximate SAL. Further theoretical analysis reveals that the proposed value smoothing technique not only helps to stabilize the training procedure of AL by controlling the trade-off between convergence rate and the upper bound of the approximation errors, but is beneficial to increase the action gap between the optimal and sub-optimal action value as well.

Zhe Zhang, Yaozhong Gan, Xiaoyang Tan

Advantage Learning (AL) seeks to increase the action gap between the optimal action and its competitors, so as to improve the robustness to estimation errors. However, the method becomes problematic when the optimal action induced by the approximated value function does not agree with the true optimal action. In this paper, we present a novel method, named clipped Advantage Learning (clipped AL), to address this issue. The method is inspired by our observation that increasing the action gap blindly for all given samples while not taking their necessities into account could accumulate more errors in the performance loss bound, leading to a slow value convergence, and to avoid that, we should adjust the advantage value adaptively. We show that our simple clipped AL operator not only enjoys fast convergence guarantee but also retains proper action gaps, hence achieving a good balance between the large action gap and the fast convergence. The feasibility and effectiveness of the proposed method are verified empirically on several RL benchmarks with promising performance.

Zhe Zhang, Karol Lasocki, Yi Yu et al.

The generation of lyrics tightly connected to accompanying melodies involves establishing a mapping between musical notes and syllables of lyrics. This process requires a deep understanding of music constraints and semantic patterns at syllable-level, word-level, and sentence-level semantic meanings. However, pre-trained language models specifically designed at the syllable level are publicly unavailable. To solve these challenging issues, we propose to exploit fine-tuning character-level language models for syllable-level lyrics generation from symbolic melody. In particular, our method endeavors to incorporate linguistic knowledge of the language model into the beam search process of a syllable-level Transformer generator network. Additionally, by exploring ChatGPT-based evaluation for generated lyrics, along with human subjective evaluation, we demonstrate that our approach enhances the coherence and correctness of the generated lyrics, eliminating the need to train expensive new language models.

Swanand Joshi, Yesu Feng, Ko-Jen Hsiao et al.

Long-lived recommender systems (RecSys) often encounter lengthy user-item interaction histories that span many years. To effectively learn long term user preferences, Large RecSys foundation models (FM) need to encode this information in pretraining. Usually, this is done by either generating a long enough sequence length to take all history sequences as input at the cost of large model input dimension or by dropping some parts of the user history to accommodate model size and latency requirements on the production serving side. In this paper, we introduce a sliding window training technique to incorporate long user history sequences during training time without increasing the model input dimension. We show the quantitative & qualitative improvements this technique brings to the RecSys FM in learning user long term preferences. We additionally show that the average quality of items in the catalog learnt in pretraining also improves.

Yuxuan Song, Zhe Zhang, Yu Pei et al.

Generative modeling of discrete variables is challenging yet crucial for applications in natural language processing and biological sequence design. We introduce the Shortlisting Model (SLM), a novel simplex-based diffusion model inspired by progressive candidate pruning. SLM operates on simplex centroids, reducing generation complexity and enhancing scalability. Additionally, SLM incorporates a flexible implementation of classifier-free guidance, enhancing unconditional generation performance. Extensive experiments on DNA promoter and enhancer design, protein design, character-level and large-vocabulary language modeling demonstrate the competitive performance and strong potential of SLM. Our code can be found at https://github.com/GenSI-THUAIR/SLM

Keyue Qiu, Yuxuan Song, Zhehuan Fan et al.

Structure-Based Drug Design (SBDD) is crucial for identifying bioactive molecules. Recent deep generative models are faced with challenges in geometric structure modeling. A major bottleneck lies in the twisted probability path of multi-modalities -- continuous 3D positions and discrete 2D topologies -- which jointly determine molecular geometries. By establishing the fact that noise schedules decide the Variational Lower Bound (VLB) for the twisted probability path, we propose VLB-Optimal Scheduling (VOS) strategy in this under-explored area, which optimizes VLB as a path integral for SBDD. Our model effectively enhances molecular geometries and interaction modeling, achieving state-of-the-art PoseBusters passing rate of 95.9% on CrossDock, more than 10% improvement upon strong baselines, while maintaining high affinities and robust intramolecular validity evaluated on held-out test set. Code is available at https://github.com/AlgoMole/MolCRAFT.

Zhe Zhang, Mingxiu Cai, Hanxiao Wang et al.

Unsupervised anomaly detection (UAD) seeks to localize the anomaly mask of an input image with respect to normal samples. Either by reconstructing normal counterparts (reconstruction-based) or by learning an image feature embedding space (embedding-based), existing approaches fundamentally rely on image-level or feature-level matching to derive anomaly scores. Often, such a matching process is inaccurate yet overlooked, leading to sub-optimal detection. To address this issue, we introduce the concept of cost filtering, borrowed from classical matching tasks, such as depth and flow estimation, into the UAD problem. We call this approach {\em CostFilter-AD}. Specifically, we first construct a matching cost volume between the input and normal samples, comprising two spatial dimensions and one matching dimension that encodes potential matches. To refine this, we propose a cost volume filtering network, guided by the input observation as an attention query across multiple feature layers, which effectively suppresses matching noise while preserving edge structures and capturing subtle anomalies. Designed as a generic post-processing plug-in, CostFilter-AD can be integrated with either reconstruction-based or embedding-based methods. Extensive experiments on MVTec-AD and VisA benchmarks validate the generic benefits of CostFilter-AD for both single- and multi-class UAD tasks. Code and models will be released at https://github.com/ZHE-SAPI/CostFilter-AD.

Boyang Wang, Weihao Zheng, Ying Wang et al.

The morphological fingerprint in the brain is capable of identifying the uniqueness of an individual. However, whether such individual patterns are present in perinatal brains, and which morphological attributes or cortical regions better characterize the individual differences of ne-onates remain unclear. In this study, we proposed a deep learning framework that projected three-dimensional spherical meshes of three morphological features (i.e., cortical thickness, mean curvature, and sulcal depth) onto two-dimensional planes through quasi-conformal mapping, and employed the ResNet18 and contrastive learning for individual identification. We used the cross-sectional structural MRI data of 682 infants, incorporating with data augmentation, to train the model and fine-tuned the parameters based on 60 infants who had longitudinal scans. The model was validated on 30 longitudinal scanned infant data, and remarkable Top1 and Top5 accuracies of 71.37% and 84.10% were achieved, respectively. The sensorimotor and visual cortices were recognized as the most contributive regions in individual identification. Moreover, the folding morphology demonstrated greater discriminative capability than the cortical thickness, which could serve as the morphological fingerprint in perinatal brains. These findings provided evidence for the emergence of morphological fingerprints in the brain at the beginning of the third trimester, which may hold promising implications for understanding the formation of in-dividual uniqueness in the brain during early development.

Zhe Zhang, Mingxiu Cai, Gaochang Wu et al.

Unsupervised anomaly detection (UAD) aims to identify image- and pixel-level anomalies using only normal training data, with wide applications such as industrial inspection and medical analysis, where anomalies are scarce due to privacy concerns and cold-start constraints. Existing methods, whether reconstruction-based (restoring normal counterparts) or embedding-based (pretrained representations), fundamentally conduct image- or feature-level matching to generate anomaly maps. Nonetheless, matching noise has been largely overlooked, limiting their detection ability. Beyond earlier focus on unimodal RGB-based UAD, recent advances expand to multimodal scenarios, e.g., RGB-3D and RGB-Text, enabled by point cloud sensing and vision-language models. Despite shared challenges, these lines remain largely isolated, hindering a comprehensive understanding and knowledge transfer. In this paper, we advocate unified UAD for both unimodal and multimodal settings in the matching perspective. Under this insight, we present Unified Cost Filtering (UCF), a generic post-hoc refinement framework for refining anomaly cost volume of any UAD model. The cost volume is constructed by matching a test sample against normal samples from the same or different modalities, followed by a learnable filtering module with multi-layer attention guidance from the test sample, mitigating matching noise and highlighting subtle anomalies. Comprehensive experiments on 22 diverse benchmarks demonstrate the efficacy of UCF in enhancing a variety of UAD methods, consistently achieving new state-of-the-art results in both unimodal (RGB) and multimodal (RGB-3D, RGB-Text) UAD scenarios. Code and models will be released at https://github.com/ZHE-SAPI/CostFilter-AD.

Yuxi Hu, Jun Zhang, Kuangyi Chen et al.

Generalizable Gaussian Splatting aims to synthesize novel views for unseen scenes without per-scene optimization. In particular, recent advancements utilize feed-forward networks to predict per-pixel Gaussian parameters, enabling high-quality synthesis from sparse input views. However, existing approaches fall short in encoding discriminative, multi-view consistent features for Gaussian predictions, which struggle to construct accurate geometry with sparse views. To address this, we propose $\mathbf{C}^{3}$-GS, a framework that enhances feature learning by incorporating context-aware, cross-dimension, and cross-scale constraints. Our architecture integrates three lightweight modules into a unified rendering pipeline, improving feature fusion and enabling photorealistic synthesis without requiring additional supervision. Extensive experiments on benchmark datasets validate that $\mathbf{C}^{3}$-GS achieves state-of-the-art rendering quality and generalization ability. Code is available at: https://github.com/YuhsiHu/C3-GS.

Tongshun Zhang, Pingping Liu, Yubing Lu et al.

Traditional Low-Light Image Enhancement (LLIE) methods primarily focus on uniform brightness adjustment, often neglecting instance-level semantic information and the inherent characteristics of different features. To address these limitations, we propose CWNet (Causal Wavelet Network), a novel architecture that leverages wavelet transforms for causal reasoning. Specifically, our approach comprises two key components: 1) Inspired by the concept of intervention in causality, we adopt a causal reasoning perspective to reveal the underlying causal relationships in low-light enhancement. From a global perspective, we employ a metric learning strategy to ensure causal embeddings adhere to causal principles, separating them from non-causal confounding factors while focusing on the invariance of causal factors. At the local level, we introduce an instance-level CLIP semantic loss to precisely maintain causal factor consistency. 2) Based on our causal analysis, we present a wavelet transform-based backbone network that effectively optimizes the recovery of frequency information, ensuring precise enhancement tailored to the specific attributes of wavelet transforms. Extensive experiments demonstrate that CWNet significantly outperforms current state-of-the-art methods across multiple datasets, showcasing its robust performance across diverse scenes. Code is available at https://github.com/bywlzts/CWNet-Causal-Wavelet-Network.

Zhe Zhang, Chunyu Wang, Weichao Qiu et al.

Occlusion is probably the biggest challenge for human pose estimation in the wild. Typical solutions often rely on intrusive sensors such as IMUs to detect occluded joints. To make the task truly unconstrained, we present AdaFuse, an adaptive multiview fusion method, which can enhance the features in occluded views by leveraging those in visible views. The core of AdaFuse is to determine the point-point correspondence between two views which we solve effectively by exploring the sparsity of the heatmap representation. We also learn an adaptive fusion weight for each camera view to reflect its feature quality in order to reduce the chance that good features are undesirably corrupted by ``bad'' views. The fusion model is trained end-to-end with the pose estimation network, and can be directly applied to new camera configurations without additional adaptation. We extensively evaluate the approach on three public datasets including Human3.6M, Total Capture and CMU Panoptic. It outperforms the state-of-the-arts on all of them. We also create a large scale synthetic dataset Occlusion-Person, which allows us to perform numerical evaluation on the occluded joints, as it provides occlusion labels for every joint in the images. The dataset and code are released at https://github.com/zhezh/adafuse-3d-human-pose.

Zhe Zhang, Chunyu Wang, Wenhu Qin et al.

We propose to estimate 3D human pose from multi-view images and a few IMUs attached at person's limbs. It operates by firstly detecting 2D poses from the two signals, and then lifting them to the 3D space. We present a geometric approach to reinforce the visual features of each pair of joints based on the IMUs. This notably improves 2D pose estimation accuracy especially when one joint is occluded. We call this approach Orientation Regularized Network (ORN). Then we lift the multi-view 2D poses to the 3D space by an Orientation Regularized Pictorial Structure Model (ORPSM) which jointly minimizes the projection error between the 3D and 2D poses, along with the discrepancy between the 3D pose and IMU orientations. The simple two-step approach reduces the error of the state-of-the-art by a large margin on a public dataset. Our code will be released at https://github.com/CHUNYUWANG/imu-human-pose-pytorch.

Anthony Hsu, Keqiu Hu, Jonathan Hung et al.

Training machine learning (ML) models on large datasets requires considerable computing power. To speed up training, it is typical to distribute training across several machines, often with specialized hardware like GPUs or TPUs. Managing a distributed training job is complex and requires dealing with resource contention, distributed configurations, monitoring, and fault tolerance. In this paper, we describe TonY, an open-source orchestrator for distributed ML jobs built at LinkedIn to address these challenges.

Kaiqiang Xiong, Rui Peng, Zhe Zhang et al.

Unsupervised Multi-View Stereo (MVS) methods have achieved promising progress recently. However, previous methods primarily depend on the photometric consistency assumption, which may suffer from two limitations: indistinguishable regions and view-dependent effects, e.g., low-textured areas and reflections. To address these issues, in this paper, we propose a new dual-level contrastive learning approach, named CL-MVSNet. Specifically, our model integrates two contrastive branches into an unsupervised MVS framework to construct additional supervisory signals. On the one hand, we present an image-level contrastive branch to guide the model to acquire more context awareness, thus leading to more complete depth estimation in indistinguishable regions. On the other hand, we exploit a scene-level contrastive branch to boost the representation ability, improving robustness to view-dependent effects. Moreover, to recover more accurate 3D geometry, we introduce an L0.5 photometric consistency loss, which encourages the model to focus more on accurate points while mitigating the gradient penalty of undesirable ones. Extensive experiments on DTU and Tanks&Temples benchmarks demonstrate that our approach achieves state-of-the-art performance among all end-to-end unsupervised MVS frameworks and outperforms its supervised counterpart by a considerable margin without fine-tuning.

Xin Gao, Li Hu, Siqi Hu et al.

Current state-of-the-art (SOTA) methods for audio-driven character animation demonstrate promising performance for scenarios primarily involving speech and singing. However, they often fall short in more complex film and television productions, which demand sophisticated elements such as nuanced character interactions, realistic body movements, and dynamic camera work. To address this long-standing challenge of achieving film-level character animation, we propose an audio-driven model, which we refere to as Wan-S2V, built upon Wan. Our model achieves significantly enhanced expressiveness and fidelity in cinematic contexts compared to existing approaches. We conducted extensive experiments, benchmarking our method against cutting-edge models such as Hunyuan-Avatar and Omnihuman. The experimental results consistently demonstrate that our approach significantly outperforms these existing solutions. Additionally, we explore the versatility of our method through its applications in long-form video generation and precise video lip-sync editing.

Zaiwei Chen, Sheng Zhang, Zhe Zhang et al.

We study the problem of solving fixed-point equations for seminorm-contractive operators and establish foundational results on the non-asymptotic behavior of iterative algorithms in both deterministic and stochastic settings. Specifically, in the deterministic setting, we prove a fixed-point theorem for seminorm-contractive operators, showing that iterates converge geometrically to the kernel of the seminorm. In the stochastic setting, we analyze the corresponding stochastic approximation (SA) algorithm under seminorm-contractive operators and Markovian noise, providing a finite-sample analysis for various stepsize choices. A benchmark for equation solving is linear systems of equations, where the convergence behavior of fixed-point iteration is closely tied to the stability of linear dynamical systems. In this special case, our results provide a complete characterization of system stability with respect to a seminorm, linking it to the solution of a Lyapunov equation in terms of positive semi-definite matrices. In the stochastic setting, we establish a finite-sample analysis for linear Markovian SA without requiring the Hurwitzness assumption. Our theoretical results offer a unified framework for deriving finite-sample bounds for various reinforcement learning algorithms in the average reward setting, including TD($λ$) for policy evaluation (which is a special case of solving a Poisson equation) and Q-learning for control.

Benjamin Kiefer, Lojze Žust, Jon Muhovič et al.

The 3rd Workshop on Maritime Computer Vision (MaCVi) 2025 addresses maritime computer vision for Unmanned Surface Vehicles (USV) and underwater. This report offers a comprehensive overview of the findings from the challenges. We provide both statistical and qualitative analyses, evaluating trends from over 700 submissions. All datasets, evaluation code, and the leaderboard are available to the public at https://macvi.org/workshop/macvi25.

Zhe Zhang, Taketo Akama

GANStrument, exploiting GANs with a pitch-invariant feature extractor and instance conditioning technique, has shown remarkable capabilities in synthesizing realistic instrument sounds. To further improve the reconstruction ability and pitch accuracy to enhance the editability of user-provided sound, we propose HyperGANStrument, which introduces a pitch-invariant hypernetwork to modulate the weights of a pre-trained GANStrument generator, given a one-shot sound as input. The hypernetwork modulation provides feedback for the generator in the reconstruction of the input sound. In addition, we take advantage of an adversarial fine-tuning scheme for the hypernetwork to improve the reconstruction fidelity and generation diversity of the generator. Experimental results show that the proposed model not only enhances the generation capability of GANStrument but also significantly improves the editability of synthesized sounds. Audio examples are available at the online demo page.