论文检索

Reinforcement Learning · Everything Else

Ju-Hyun Kim, Seungki Min

We propose a sample-efficient Q-learning algorithm for reinforcement learning with the Conditional Value-at-Risk (CVaR) objective. Our algorithm is built upon predictive tail value function, a novel formulation of risk-sensitive action value, that admits a recursive structure as in the conventional risk-neutral Bellman equation. This structure enables the Q-learning algorithm to utilize the entire set of sample trajectories rather than relying only on worst-case outcomes, enhancing the sample efficiency. We further derive a Bellman optimality equation and a policy improvement theorem, which provide theoretical foundations of our algorithm and remedy inconsistencies that have existed in the literature. Empirical results demonstrate that our method consistently improves CVaR performance while maintaining stable and interpretable learning dynamics.

Computer Vision · Everything Else

Jiachen Qian, Hongye Yang, Youtian Lin, tianhao zhao, Feihu Zhang, Yao Yao, Hengshuang Zhao

3D anime character generation has become increasingly important in digital entertainment, including animation production, virtual reality, gaming, and virtual influencers. Unlike realistic human modeling, anime-style characters require exaggerated proportions, stylized surface details, and artistically consistent garments, posing unique challenges for automated 3D generation. Previous approaches for 3D anime character generation often suffer from low mesh quality and blurry textures, and they typically do not provide corresponding skeletons, limiting their usability in animation. In this work, we present a novel framework for high-quality 3D anime character generation that overcomes these limitations by combining the expressive power of the Skinned Multi-Person Linear (SMPL) model with precise garment generation. Our approach extends the Anime-SMPL model to better capture the distinct features of anime characters, enabling unified skeleton generation and blendshape-based facial expression control. This results in fully animation-ready 3D characters with expressive faces, bodies, and garments. To complement the body model, we introduce a body-aligned component-wise garments generation pipeline (including hairstyles, upper garments, lower garments, and accessories), which models garments as structured components aligned with body geometry. Furthermore, our method produces high-quality skin and facial textures, as well as detailed garment textures, enhancing the visual fidelity of the generated characters. Experimental results demonstrate that our framework significantly outperforms baseline methods in terms of mesh quality, texture clarity, and garment-body alignment, making it suitable for a wide range of applications in anime content creation and interactive media.

Computer Vision · Everything Else

Chenliang Zhou, Fangcheng Zhong, Weihao Xia, Albert Miao, Canberk Baykal, Cengiz Oztireli

We introduce the *Quartet of Diffusions*, a structure-aware point cloud generation framework that explicitly models part composition and symmetry. Unlike prior methods that treat shape generation as a holistic process or only support part composition, our approach leverages four coordinated diffusion models to learn distributions of global shape latents, symmetries, semantic parts, and their spatial assembly. This structured pipeline ensures guaranteed symmetry, coherent part placement, and diverse, high-quality outputs. By disentangling the generative process into interpretable components, our method supports fine-grained control over shape attributes, enabling targeted manipulation of individual parts while preserving global consistency. A central global latent further reinforces structural coherence across assembled parts. Our experiments show that the Quartet achieves state-of-the-art performance. To our best knowledge, this is the first 3D point cloud generation framework that fully integrates and enforces both symmetry and part priors throughout the generative process.

Jan-Niklas Dihlmann, Mark Boss, Simon Donné, Andreas Engelhardt, Hendrik Lensch, Varun Jampani

Reconstructing 3D assets from images has long required separate pipelines for geometry reconstruction, material estimation, and illumination recovery, each with distinct limitations and computational overhead. We present MIDR-3D, the first unified end-to-end pipeline that simultaneously reconstructs complete 3D geometry, spatially-varying physically-based materials, and environment illumination from sparse multi-view images in under one second. Our key insight is that multi-view constraints can dramatically improve material and illumination disentanglement, a problem that remains fundamentally ill-posed for single image methods. Key to our approach is the fusion of the multi-view input via a transformer cross-conditioning architecture, followed by a novel unified two path prediction strategy. The first path predicts the object’s structure and appearance, while the second path predicts the environment illumination from image background or object reflections. This combined with a differentiable Monte Carlo multiple importance sampling renderer, creates an optimal illumination disentanglement training pipeline. Further with our mixed-domain training protocol, combining synthetic PBR datasets with real-world RGB captures, we establish generalizable results across geometry, material accuracy, and illumination quality. By unifying previously separate reconstruction tasks into a single feed-forward pass, we enable near-instantaneous generation of complete, relightable 3D assets.

Deep Learning · Generative Models and Autoencoders

Yuhao Shen, Junyi Shen, Quan Kong, Tianyu Liu, Yao Lu, Cong Wang

Speculative decoding (SD) has emerged as a promising technique to accelerate LLM inference by employing a small, efficient draft model to propose draft tokens in advance, and subsequently validating them in parallel with the large target model. However, the existing SD methods still remain fundamentally constrained by their serialized execution, which inevitably causes mutual waiting bubbles between the draft and target models. To address this critical challenge, we draw inspiration from sophisticated branch prediction mechanisms in modern processors and propose a novel framework, \textbf{SpecBranch}, to fully unlock branch parallelism in SD. Specifically, we first conduct an in-depth analysis of the potential of branch parallelism in SD, and recognize that the key challenge lies in the intricate trade-offs between parallelization and token rollback. Based on this analysis, we introduce parallel speculative branches to preemptively hedge against likely rejections. Meanwhile, to significantly enhance parallelism, we jointly orchestrate adaptive draft lengths with a hybrid combination of the implicit draft model confidence and explicit reusing of target model features. Extensive experiments conducted across various models and benchmarks show that \textbf{SpecBranch} achieves impressive speedups of over \textbf{1.8}$\times \sim$ \textbf{4.5}$\times$ against the standard auto-regressive decoding and reduces rollback tokens by \textbf{50}\% for poorly aligned models, while maintaining an identical sampling distribution. Our code is available at \url{https://github.com/Sylvan820/Specbranch}.

Computer Vision · Vision Models & Multimodal

Zongxia Li, Wenhao Yu, Chengsong Huang, Zhenwen Liang, Rui Liu, Fuxiao Liu, Jingxi Chen, Dian Yu, Jordan Boyd-Graber, Haitao Mi 等

Vision-Language Models (VLMs) often suffer from visual hallucinations -- generating things that are not consistent with visual inputs -- and language shortcuts, where they skip the visual part and just rely on text priors. These issues arise because most post-training methods for VLMs rely on simple verifiable answer matching and supervise only final outputs, leaving intermediate visual reasoning without explicit guidance. As a result, VLMs receive sparse visual signals and often learn to prioritize language-based reasoning over visual perception. To mitigate this, some existing methods add visual supervision using human annotations or distilled labels from external large models. However, human annotations are labor-intensive and costly, and external signals can introduce high latency cost. In this paper, we introduce Vision-SR1, a three-stage self-rewarding reinforcement learning method that improves visual reasoning without relying on external visual supervision. Vision-SR1 decomposes VLM reasoning into two components: visual reasoning and language reasoning, where the model is first prompted to produce self-contained visual descriptions sufficient to answer the question without referring back to the input image, before jointly optimizing both visual and language reasoning through our multi-reward loss objective. To validate this self-containment, the same VLM model is re-prompted to perform language reasoning using only the generated visual reasoning as input to compute visual reward. The final reward is computed through a decoupled reward-advantage framework, where visual reward and language reasoning reward each have their advantages, log probabilities, and KL divergence calculated separately. This decoupling enables more fine-grained reward computation by preventing the entanglement of heterogeneous reward signals. Our experiments show that Vision-SR1 improves visual reasoning, mitigates visual hallucinations, and reduces reliance on language shortcuts across diverse vision-language tasks, while being more efficient than methods that rely on external visual reward models, which require additional GPUs to host. In contrast, Vision-SR1 introduces no extra GPU overhead beyond that of standard training.

Computer Vision · Image and Video Generation

Joshua Tian Jin Tee, Hee Suk Yoon, Sunjae Yoon, Tri Ton, Chang Yoo

Classifier-Free Guidance (CFG) improves diffusion sampling by encouraging conditional generations while discouraging unconditional ones. Existing preference alignment methods, however, focus only on positive preference pairs, limiting their ability to actively suppress undesirable outputs. Diffusion Negative Preference Optimization (Diff-NPO) approaches this limitation by introducing a separate negative model trained with inverted labels, allowing it to capture signals for suppressing undesirable generations. However, this design comes with two key drawbacks. First, maintaining two distinct models throughout training and inference substantially increases computational cost, making the approach less practical. Second, at inference time, Diff-NPO relies on weight merging between the positive and negative models, a process that dilutes the learned negative alignment and undermines its effectiveness. To overcome these issues, we introduce Diff-SNPO, a single-network framework that jointly learns from both positive and negative preferences. Our method employs a bounded preference objective to prevent winner-likelihood collapse, ensuring stable optimization. Diff-SNPO delivers strong alignment performance with significantly lower computational overhead, showing that explicit negative preference modeling can be simple, stable, and efficient within a unified diffusion framework. Code will be released.

Remi Cadene, Simon Alibert, Francesco Capuano, Michel Aractingi, Adil Zouitine, Pepijn Kooijmans, Jade Choghari, Martino Russi, Caroline Pascal, Steven Palma 等

Robotics is undergoing a significant transformation powered by advances in high-level control techniques based on machine learning, giving rise to the field of robot learning. Recent progress in robot learning has been accelerated by the increasing availability of affordable teleoperation systems, large-scale openly available datasets, and scalable learning-based methods. However, development in the field of robot learning is often slowed by fragmented, closed-source tools designed to only address specific sub-components within the robotics stack. In this paper, we present lerobot, an open-source library that integrates across the entire robotics stack, from low-level middleware communication for motor controls to large-scale dataset collection, storage and streaming. The library is designed with a strong focus on real-world robotics, supporting accessible hardware platforms while remaining extensible to new embodiments. It also supports efficient implementations for various state-of-the-art robot learning algorithms from multiple prominent paradigms, as well as a generalized asynchronous inference stack. Unlike traditional pipelines which heavily rely on hand-crafted techniques, lerobot emphasizes scalable learning approaches that improve directly with more data and compute. Designed for accessibility, scalability, and openness, lerobot lowers the barrier to entry for researchers and practitioners to robotics while providing a platform for reproducible, state-of-the-art robot learning.

Applications · Genetics, Cell Biology, Health, etc

Zelin Liu, Dongdong Chen, Yusong Sun, Yuqi Hu, Huang Jie, Sicheng Dong, Xu Han, Hongmei Yi, Qiyuan Bao, Lichi Zhang

Segmentation in pathology image requires handling out-of-domain tissue morphologies and new pathologies beyond training distributions, where traditional closed-set segmentation approaches fail to generalize. Reasoning segmentation enables zero-shot generalization via prompting with text queries. However, existing reasoning segmentation models face three barriers when applied to pathology: (1) the vision encoder lack pathology-specific knowledge and robustness to staining variations, (2) the large language model (LLM) backbone for reasoning fails to identify whether it has gathered sufficient semantic context to trigger the segmentation output, and (3) no reasoning segmentation benchmarks and datasets exist for pathology analysis. Consequently, we introduce PathChat-SegR1, a reasoning segmentation model built upon pathology-specific vision encoders trained with a novel stain-invariant self-distillation for robust pathology image representations. Moreover, we propose Segmentation-Optimized GRPO (SO-GRPO), a reinforcement learning method specifically for reasoning segmentation that learns to determine optimal segmentation timing based on accumulated reasoning context. Finally, we construct a pathology-specific reasoning segmentation benchmark of 118,667 triplets of pathology image, ground-truth mask, query, and reasoning chain including both public and private pathology images. Zero-shot evaluation on pathology images with out-of-domain morphologies/pathologies shows 61\% improvement over state-of-the-art segmentation models.

Applications · Language, Speech and Dialog

Yefan Zhou, Austin Xu, Yilun Zhou, Janvijay Singh, Jiang Gui, Shafiq Joty

Recent advances have shown that scaling test-time computation enables large language models (LLMs) to solve increasingly complex problems across diverse domains. One effective paradigm for test-time scaling (TTS) involves LLM generators producing multiple solution candidates, with LLM verifiers assessing the correctness of these candidates without reference answers. In this paper, we study generative verifiers, which perform verification by generating chain-of-thought (CoT) reasoning followed by a binary verdict. We systematically analyze verification dynamics across three dimensions -- problem difficulty, generator capability, and verifier generation capability -- through empirical studies on 12 benchmarks across mathematical reasoning, knowledge, and natural language reasoning tasks using 14 open-source models (2B to 72B parameter range) and GPT-4o. Our experiments reveal three key findings about verification effectiveness: (1) Easy problems allow verifiers to more reliably certify correct responses; (2) Weak generators produce errors that are easier to detect than strong generators; (3) Verification ability is generally correlated with the verifier's own problem-solving capability, but this relationship varies with problem difficulty. These findings reveal opportunities for optimizing basic verification strategies in TTS applications. First, given the same verifier, some weak generators can nearly match stronger ones in post-verification TTS performance (e.g., the Gemma2-9B to Gemma2-27B performance gap shrinks by 75.7%). Second, we identify cases where strong verifiers offer limited advantages over weak ones, as both fail to provide meaningful verification gains, suggesting that verifier scaling alone cannot overcome fundamental verification challenges.

Computer Vision · Image and Video Generation

Jiangshan Wang, Kang Zhao, Jiayi Guo, Jiayu Wang, Hang Guo, Chenyang Zhu, Xiu Li, Xiangyu Yue

High computational costs and slow inference hinder the practical application of video generation models. While prior works accelerate the generation process through feature caching, they often suffer from notable quality degradation. In this work, we reveal that this issue arises from their inability to distinguish truly redundant features, which leads to the unintended skipping of computations on important features. To address this, we propose \textbf{PreciseCache}, a plug-and-play framework that precisely detects and skips truly redundant computations, thereby accelerating inference without sacrificing quality. Specifically, PreciseCache contains two components: LFCache for step-wise caching and BlockCache for block-wise caching. For LFCache, we compute the Low-Frequency Difference (LFD) between the prediction features of the current step and those from the previous cached step. Empirically, we observe that LFD serves as an effective measure of step-wise redundancy, accurately detecting highly redundant steps whose computation can be skipped through reusing cached features. To further accelerate generation within each non-skipped step, we propose BlockCache, which precisely detects and skips redundant computations at the block level within the network. Extensive experiments on various backbones demonstrate the effectiveness of our PreciseCache, which achieves an average of $2.6\times$ speedup without noticeable quality loss. Source code will be released.

Social Aspects · Privacy-preserving Statistics and Machine Learning

Nikita Kalinin, Ryan McKenna, Jalaj Upadhyay, Christoph Lampert

Matrix factorization mechanisms for differentially private training have emerged as a promising approach to improve model utility under privacy constraints. In practical settings, models are typically trained over multiple epochs, requiring matrix factorizations that account for repeated participation. Existing theoretical upper and lower bounds on multi-epoch factorization error leave a significant gap. In this work, we introduce a new explicit factorization method, Banded Inverse Square Root (BISR), which imposes a banded structure on the inverse correlation matrix. This factorization enables us to derive an explicit and tight characterization of the multi-epoch error. We further prove that BISR achieves asymptotically optimal error by matching the upper and lower bounds. Empirically, BISR performs on par with the state of the art factorization methods, while being simpler to implement, computationally efficient, and easier to analyze.

Fangyu Lei, Jinxiang Meng, Yiming Huang, Junjie zhao, Yitong Zhang, Jianwen Luo, Xin Zou, Ruiyi Yang, Wenbo Shi, Yan Gao 等

Real-world enterprise data intelligence workflows encompass data engineering that turns raw sources into analytical-ready tables and data analysis that convert those tables into decision-oriented insights. We introduce DAComp, a benchmark of 210 tasks that mirrors these complex workflows. Data engineering (DE) tasks require repository-level engineering on industrial schemas, including designing and building multi-stage SQL pipelines from scratch and evolving existing systems under evolving requirements. Data analysis (DA) tasks pose open-ended business problems that demand strategic planning, exploratory analysis through iterative coding, interpretation of intermediate results, and the synthesis of actionable recommendations. Engineering tasks are scored through execution-based, multi-metric evaluation. Open-ended tasks are assessed by a reliable, experimentally validated LLM-judge, which is guided by hierarchical, meticulously crafted rubrics. Our experiments reveal that even state-of-the-art agents falter on DAComp. Performance on DE tasks is particularly low, with success rates under 20\%, exposing a critical bottleneck in holistic pipeline orchestration, not merely code generation. Scores on DA tasks also average below 40\%, highlighting profound deficiencies in open-ended reasoning and demonstrating that engineering and analysis are distinct capabilities. By clearly diagnosing these limitations, DAComp provides a rigorous and realistic testbed to drive the development of truly capable autonomous data agents for enterprise settings. Our data and code are available at \url{da-comp.github.io}.

Applications · Climate

Jiawei Chen, Weiqi Chen, Rong Hu, Peiyuan Liu, Haifan Zhang, Liang Sun

Solving primitive equations is essential for accurate weather forecasting. However, traditional numerical weather prediction (NWP) methods often incorporate various simplifications that limit their effectiveness in parameterizing unresolved physical processes. Meanwhile, existing deep learning-based models mostly focus on pure data-driven paradigms, overlooking the fundamental physical principles that govern atmospheric dynamics. To address these challenges, we present DeepPrim, a novel 3D \underline{deep} weather forecaster designed to learn \underline{prim}itive equations of the Earth’s atmosphere. Specifically, DeepPrim aims at accurately modeling 3D atmospheric motion through Navier-Stokes equation in pressure coordinates, and effectively capturing the interactions between the solved advection and key weather variables (e.g., temperature and water vapor) through corresponding equations. By seamlessly integrating fundamental atmospheric physics with advanced data-driven techniques, our model effectively approximates complicated physical processes without relying on empirical simplifications. Experimentally, DeepPrim achieves impressive performance in both short-term global and regional weather forecasting tasks, and exhibits the superior capacity to capture 3D atmospheric dynamics. It is now deployed as part of the Baguan weather forecasting system, especially specializing in short-term forecasting. The code is available at https://github.com/DAMO-DI-ML/DeepPrim.

Hai Huang, Yann LeCun, Randall Balestriero

Large Language Model (LLM) pretraining, finetuning, and evaluation rely on input-space reconstruction and generative capabilities. Yet, it has been observed in vision that embedding-space training objectives, e.g., with Joint Embedding Predictive Architectures (JEPAs), are far superior to their input-space counterpart. That mismatch in how training is achieved between language and vision opens up a natural question: {\em can language training methods learn a few tricks from the vision ones?} The lack of JEPA-style LLM is a testimony of the challenge in designing such objectives for language. In this work, we propose a first step in that direction where we develop LLM-JEPA, a JEPA based solution for LLMs applicable both to finetuning and pretraining. Thus far, LLM-JEPA is able to outperform the standard LLM training objectives by a significant margin across models, all while being robust to overfiting. Those findings are observed across numerous datasets (NL-RX, GSM8K, Spider, RottenTomatoes) and various models from the Llama3, OpenELM, Gemma2 and Olmo families. Code: \url{https://github.com/galilai-group/llm-jepa}.

Zhiping Xiao, Junyu Luo, Hang Zhou, Yusheng Zhao, Xiao Luo, Pengyun Wang, Wei Ju, Siyu Heng, Ming Zhang

Cross-modal retrieval is a fundamental task that aims to learn semantic correspondences across different data modalities, such as visual and textual modalities. Unsupervised hashing methods can efficiently manage large-scale data and can be effectively applied to cross-modal retrieval studies.However, existing methods typically fail to fully exploit the hierarchical semantic structure within text and image data, where instances naturally organize into multi-level communities of varying granularity. Moreover, the commonly-used direct modal alignment cannot effectively bridge the semantic gap between these two modalities. To address these issues, we introduce a novel Hierarchical Encoding Tree with Modality Mixup (HINT) method, which achieves effective cross-modal retrieval by extracting hierarchical cross-modal relations. HINT constructs a cross-modal encoding tree guided by hierarchical structural entropy and generates proxy samples of text and image modalities for each instance from the encoding tree. Through the curriculum-based mixup of proxy samples, HINT achieves progressive modal alignment and effective cross-modal retrieval. We also conduct cross-modal consistency learning to achieve global-view semantic alignment between text and image representations. Extensive experiments on a range of cross-modal retrieval datasets demonstrate the superiority of HINT over state-of-the-art methods.

Computer Vision · Image and Video Generation

Yuan Zeng, Yujia Shi, Yuhao Yang, Dongxia Liu, Zongqing Lu, Wenming Yang, Qingmin Liao

Human image animation aims to generate a video from a static reference image, guided by pose information extracted from a driving video. Existing approaches often rely on pose estimators to extract intermediate representations, but such signals are prone to errors under occlusion or complex poses. Building on these observations, we present DirectAnimator, a framework that bypasses pose extraction and directly learns from raw driving videos. We introduce a Driving Cue Triplet consisting of pose, face, and location cues that captures motion, expression, and alignment in a semantically rich yet stable form, and we fuse them through a CueFusion DiT block for reliable control during denoising. To make learning dependable when the driving and reference identities differ, we devise a Same2X training strategy that aligns cross-ID features with those learned from same-ID data, regularizing optimization and accelerating convergence. Extensive experiments demonstrate that DirectAnimator attains state-of-the-art visual quality and identity preservation while remaining robust to occlusions and complex articulation, and it does so with fewer computational resources. Our project page is at https://directanimator.github.io/.

Kuan Li, Zhongwang Zhang, Huifeng Yin, Rui Ye, Yida Zhao, Liwen Zhang, Litu Ou, Ding-Chu Zhang, Xixi Wu, Xinmiao Yu 等

To significantly advance the capabilities of open-source web agents, we present WebSailor-V2, a complete post-training pipeline encompassing data construction, Supervised Fine-Tuning (SFT), and Reinforcement Learning (RL). Our methodology features two key innovations: (1) On the data front, we developed SailorFog-QA-2, a novel dataset built from a densely interconnected knowledge graph that introduces a wide variety of uncertainties beyond simple obfuscation, fostering more sophisticated reasoning. (2) For training, we engineered a dual-environment RL framework, combining a high-fidelity simulator for rapid, low-cost algorithmic iteration with a robust, managed real-world environment for stable final policy training, all integrated within a symbiotic data-policy feedback loop. Trained on the Qwen3-30B-A3B model, WebSailor-V2 achieves state-of-the-art results, scoring 35.3 on BrowseComp-EN, 44.1 on BrowseComp-ZH, and 30.6 on Humanity's Last Exam (HLE). Notably, our 30B-A3B MOE agent significantly outperforms all existing open-source agents and surpasses even the 671B DeepSeek-V3.1, demonstrating performance competitive with leading proprietary systems.

Computer Vision · Segmentation

Tianpei Zou, Guo Yu, Ya Wu, Fan Lu, Eric Zhongcong Xu, Zhang Bo, Ziqiao Wang, Sanqing Qu, Guang Chen

Open-set Test-time Adaptation (OSTTA) has been introduced to address the challenges of both online model optimization and open-set recognition. Despite the demonstrated success of OSTTA methodologies in 2D image recognition, their application to 3D point cloud semantic segmentation is still hindered by the complexities of point cloud data, particularly the imbalance between known (in-distribution, ID) and unknown (out-of-distribution, OOD) data, where known samples dominate and unknown instances are often sparse or even absent. In this paper, we propose a simple yet effective strategy, termed Geometry-guided Out-of-Distribution Modeling (GOOD), specifically designed to address OSTTA for 3D point cloud semantic segmentation. Technically, we first leverage geometric priors to cluster the point cloud into superpoints, thereby mitigating the numerical disparity between individual points and providing a more structured data representation. Then, we introduce a novel confidence metric to effectively distinguish between known and unknown superpoints. Additionally, prototype-based representations are integrated to enhance the discrimination between ID and OOD regions, facilitating robust segmentation. We validate the efficacy of GOOD across four benchmark datasets. Remarkably, on the Synth4D to SemanticKITTI task, GOOD outperforms HGL by 1.93%, 8.99%, and 7.91% in mIoU, AUROC, and FPR95, respectively.

Xueyao Zhang, Chaoren Wang, Huan Liao, Ziniu Li, Yuancheng Wang, Li Wang, Dongya Jia, Yuanzhe Chen, Xiulin LI, Zhuo Chen 等

Aligning large generative models with human feedback is a critical challenge. In speech synthesis, this is particularly pronounced due to the lack of a large-scale human preference dataset, which hinders the development of models that truly align with human perception. To address this, we introduce ***SpeechJudge***, a comprehensive suite comprising a dataset, a benchmark, and a reward model centered on naturalness—one of the most fundamental subjective metrics for speech synthesis. First, we present ***SpeechJudge-Data***, a large-scale human feedback corpus of 99k speech pairs. The dataset is constructed using a diverse set of advanced zero-shot text-to-speech (TTS) models across diverse speech styles and multiple languages, with human annotations for both intelligibility and naturalness preference. From this, we establish ***SpeechJudge-Eval***, a challenging benchmark for speech naturalness judgment. Our evaluation reveals that existing metrics and AudioLLMs struggle with this task; the best-performing model, Gemini-2.5-Flash, achieves less than 70% agreement with human judgment, highlighting a significant gap for improvement. To bridge this gap, we develop ***SpeechJudge-GRM***, a generative reward model (GRM) based on Qwen2.5-Omni-7B. It is trained on SpeechJudge-Data via a two-stage post-training process: Supervised Fine-Tuning (SFT) with Chain-of-Thought rationales followed by Reinforcement Learning (RL) with GRPO on challenging cases. On the SpeechJudge-Eval benchmark, the proposed SpeechJudge-GRM demonstrates superior performance, achieving 77.2% accuracy (and 79.4% after inference-time scaling @10) compared to a classic Bradley-Terry reward model (72.7%). Furthermore, SpeechJudge-GRM can be also employed as a reward function during the post-training of speech generation models to facilitate their alignment with human preferences.