Large Multimodal Models (LMMs) have recently demonstrated remarkable visual understanding performance on both vision-language and vision-centric tasks. However, they often fall short in integrating advanced, task-specific capabilities for compositional reasoning, which hinders their progress toward truly competent general vision models. To address this, we present a unified visual reasoning mechanism that enables LMMs to solve complicated compositional problems by leveraging their intrinsic capabilities (e.g. grounding and visual understanding capabilities). Different from the previous shortcut learning mechanism, our approach introduces a human-like understanding-thinking-answering process, allowing the model to complete all steps in a single pass forwarding without the need for multiple inferences or external tools. This design bridges the gap between foundational visual capabilities and general question answering, encouraging LMMs to generate faithful and traceable responses for complex visual reasoning. Meanwhile, we curate 334K visual instruction samples covering both general scenes and text-rich scenes and involving multiple foundational visual capabilities. Our trained model, Griffon-R, has the ability of end-to-end automatic understanding, self-thinking, and reasoning answers. Comprehensive experiments show that Griffon-R not only achieves advancing performance on complex visual reasoning benchmarks including VSR and CLEVR, but also enhances multimodal capabilities across various benchmarks like MMBench and ScienceQA. Data, models, and codes will be release at https://github.com/jefferyZhan/Griffon/tree/master/Griffon-R soon.
Occupancy prediction aims to estimate the 3D spatial distribution of occupied regions along with their corresponding semantic labels. Existing vision-based methods perform well on daytime benchmarks but struggle in nighttime scenarios due to limited visibility and challenging lighting conditions. To address these challenges, we propose \textbf{LIAR}, a novel framework that learns illumination-affined representations. LIAR first introduces Selective Low-light Image Enhancement (SLLIE), which leverages the illumination priors from daytime scenes to adaptively determine whether a nighttime image is genuinely dark or sufficiently well-lit, enabling more targeted global enhancement. Building on the illumination maps generated by SLLIE, LIAR further incorporates two illumination-aware components: 2D Illumination-guided Sampling (2D-IGS) and 3D Illumination-driven Projection (3D-IDP), to respectively tackle local underexposure and overexposure. Specifically, 2D-IGS modulates feature sampling positions according to illumination maps, assigning larger offsets to darker regions and smaller ones to brighter regions, thereby alleviating feature degradation in underexposed areas. Subsequently, 3D-IDP enhances semantic understanding in overexposed regions by constructing illumination intensity fields and supplying refined residual queries to the BEV context refinement process. Extensive experiments on both real and synthetic datasets demonstrate the superior performance of LIAR under challenging nighttime scenarios. The source code and pretrained models are available \href{https://github.com/yanzq95/LIAR}{here}.
Referring Multi-object tracking (RMOT) is an important research field in computer vision. Its task form is to guide the models to track the objects that conform to the language instruction. However, the RMOT task commonly requires clear language instructions, such methods often fail to work when complex language instructions with reasoning characteristics appear. In this work, we propose a new task, called Reasoning-based Multi-Object Tracking (ReaMOT). ReaMOT is a more challenging task that requires accurate reasoning about objects that match the language instruction with reasoning characteristic and tracking the objects' trajectories. To advance the ReaMOT task and evaluate the reasoning capabilities of tracking models, we construct ReaMOT Challenge, a reasoning-based multi-object tracking benchmark built upon 12 datasets. Specifically, it comprises 1,156 language instructions with reasoning characteristic, 423,359 image-language pairs, and 869 diverse scenes, which is divided into three levels of reasoning difficulty. In addition, we propose a set of evaluation metrics tailored for the ReaMOT task. Furthermore, we propose ReaTrack, a training-free framework for reasoning-based multi-object tracking based on large vision-language models (LVLM) and SAM2, as a baseline for the ReaMOT task. Extensive experiments on the ReaMOT Challenge benchmark demonstrate the effectiveness of our ReaTrack framework.
Precise control over language model generation is vital for ensuring both safety and reliability. Although prompt engineering and steering are commonly used to intervene in model behaviors, the vast number of parameters in models often results in highly intertwined internal representations. This interdependency can limit control precision and sometimes lead to unintended side effects. Recent research has explored the use of sparse autoencoders (SAE) to disentangle knowledge in high-dimensional spaces for steering. However, these applications have been limited to toy tasks owing to the nontrivial issue of locating atomic knowledge components. In this paper, we propose Steering Target Atoms (STA), a novel method that isolates and manipulates disentangled knowledge components to enhance safety. Comprehensive experiments demonstrate the effectiveness of our approach. Further analysis reveals that steering exhibits superior robustness and flexibility, particularly in adversarial scenarios. We also apply the steering strategy to the large reasoning model, confirming its effectiveness in precise reasoning control.
Modern reasoning models, such as OpenAI's o1 and DeepSeek-R1, exhibit impressive problem-solving capabilities but suffer from critical inefficiencies: high inference latency, excessive computational resource consumption, and a tendency toward overthinking -- generating verbose chains of thought (CoT) laden with redundant tokens that contribute minimally to the final answer. To address these issues, we propose Conditional Token Selection (CTS), a token-level compression framework with a flexible and variable compression ratio that identifies and preserves only the most essential tokens in CoT. CTS evaluates each token's contribution to deriving correct answers using conditional importance scoring, then trains models on compressed CoT. Extensive experiments demonstrate that CTS effectively compresses long CoT while maintaining strong reasoning performance. Notably, on the GPQA benchmark, Qwen2.5-14B-Instruct trained with CTS achieves a 9.1% accuracy improvement with 13.2% fewer reasoning tokens (13% training token reduction). Further reducing training tokens by 42% incurs only a marginal 5% accuracy drop while yielding a 75.8% reduction in reasoning tokens, highlighting the prevalence of redundancy in existing CoT.
Recent large reasoning models (LRMs) have demonstrated strong reasoning capabilities through reinforcement learning (RL). These improvements have primarily been observed within the short-context reasoning tasks. In contrast, extending LRMs to effectively process and reason on long-context inputs via RL remains a critical unsolved challenge. To bridge this gap, we first formalize the paradigm of long-context reasoning RL, and identify key challenges in suboptimal training efficiency and unstable optimization process. To address these issues, we propose QwenLong-L1, a framework that adapts short-context LRMs to long-context scenarios via progressive context scaling. Specifically, we utilize a warm-up supervised fine-tuning (SFT) stage to establish a robust initial policy, followed by a curriculum-guided phased RL technique to stabilize the policy evolution, and enhanced with a difficulty-aware retrospective sampling strategy to incentivize the policy exploration. Experiments on seven long-context document question-answering benchmarks demonstrate that QwenLong-L1-32B outperforms flagship LRMs like OpenAI-o3-mini and Qwen3-235B-A22B, achieving performance on par with Claude-3.7-Sonnet-Thinking, demonstrating leading performance among state-of-the-art LRMs. This work advances the development of practical long-context LRMs capable of robust reasoning across information-intensive environments.
In this paper, we propose a simple and efficient method for value model training on long-context reasoning traces. Compared to existing process reward models (PRMs), our method does not require a fine-grained notion of "step," which is difficult to define for long-context reasoning models. By collecting a dataset of 2.5 million reasoning traces, we train a 1.5B token-level value model and apply it to DeepSeek models for improved performance with test-time compute scaling. We find that block-wise value-guided search (VGS) with a final weighted majority vote achieves better test-time scaling than standard methods such as majority voting or best-of-n. With an inference budget of 64 generations, VGS with DeepSeek-R1-Distill-1.5B achieves an average accuracy of 45.7% across four competition math benchmarks (AIME 2024 & 2025, HMMT Feb 2024 & 2025), reaching parity with o3-mini-medium. Moreover, VGS significantly reduces the inference FLOPs required to achieve the same performance of majority voting. Our dataset, model and codebase are open-sourced.
Large language models excel at complex tasks by breaking down problems into structured reasoning steps. However, reasoning traces often extend beyond reaching a correct answer, causing wasted computation, reduced readability, and hallucinations. To address this, we introduce a novel hyperparameter-free conciseness score used as a reward signal within a reinforcement learning framework to guide models toward generating correct and concise reasoning traces. This score is evaluated by a large language model acting as a judge, enabling dynamic, context-aware feedback beyond simple token length. Our method achieves state-of-the-art efficiency-accuracy trade-offs on the MATH dataset, reducing token usage by up to 31x on simple problems while improving accuracy by 7%, and on the hardest problems, it outperforms full reasoning by +7.5% accuracy with up to 3.6x fewer tokens. On TheoremQA, our method improves accuracy by +2.2% using 12.5x fewer tokens. We also conduct ablation studies on the judge model, reward composition, and problem difficulty, showing that our method dynamically adapts reasoning length based on problem difficulty and benefits significantly from stronger judges. The code, model weights, and datasets are open-sourced at https://github.com/RazvanDu/ConciseRL.
The ViDoRe Benchmark V1 was approaching saturation with top models exceeding 90% nDCG@5, limiting its ability to discern improvements. ViDoRe Benchmark V2 introduces realistic, challenging retrieval scenarios via blind contextual querying, long and cross-document queries, and a hybrid synthetic and human-in-the-loop query generation process. It comprises four diverse, multilingual datasets and provides clear evaluation instructions. Initial results demonstrate substantial room for advancement and highlight insights on model generalization and multilingual capability. This benchmark is designed as a living resource, inviting community contributions to maintain relevance through future evaluations.
Robust tooling and publicly available pre-trained models have helped drive recent advances in mechanistic interpretability for language models. However, similar progress in vision mechanistic interpretability has been hindered by the lack of accessible frameworks and pre-trained weights. We present Prisma (Access the codebase here: https://github.com/Prisma-Multimodal/ViT-Prisma), an open-source framework designed to accelerate vision mechanistic interpretability research, providing a unified toolkit for accessing 75+ vision and video transformers; support for sparse autoencoder (SAE), transcoder, and crosscoder training; a suite of 80+ pre-trained SAE weights; activation caching, circuit analysis tools, and visualization tools; and educational resources. Our analysis reveals surprising findings, including that effective vision SAEs can exhibit substantially lower sparsity patterns than language SAEs, and that in some instances, SAE reconstructions can decrease model loss. Prisma enables new research directions for understanding vision model internals while lowering barriers to entry in this emerging field.
Visual Language Models require substantial computational resources for inference due to the additional input tokens needed to represent visual information. However, these visual tokens often contain redundant and unimportant information, resulting in an unnecessarily high number of tokens. To address this, we introduce PACT, a method that reduces inference time and memory usage by pruning irrelevant tokens and merging visually redundant ones at an early layer of the language model. Our approach uses a novel importance metric to identify unimportant tokens without relying on attention scores, making it compatible with FlashAttention. We also propose a novel clustering algorithm, called Distance Bounded Density Peak Clustering, which efficiently clusters visual tokens while constraining the distances between elements within a cluster by a predefined threshold. We demonstrate the effectiveness of PACT through extensive experiments.
Mamba-based vision models have gained extensive attention as a result of being computationally more efficient than attention-based models. However, spatial redundancy still exists in these models, represented by token and block redundancy. For token redundancy, we analytically find that early token pruning methods will result in inconsistency between training and inference or introduce extra computation for inference. Therefore, we customize token pruning to fit the Mamba structure by rearranging the pruned sequence before feeding it into the next Mamba block. For block redundancy, we allow each image to select SSM blocks dynamically based on an empirical observation that the inference speed of Mamba-based vision models is largely affected by the number of SSM blocks. Our proposed method, Dynamic Vision Mamba (DyVM), effectively reduces FLOPs with minor performance drops. We achieve a reduction of 35.2\% FLOPs with only a loss of accuracy of 1.7\% on Vim-S. It also generalizes well across different Mamba vision model architectures and different vision tasks. Our code will be made public.
While state-of-the-art language models achieve impressive results through next-token prediction, they have inherent limitations such as the inability to revise already generated tokens. This has prompted exploration of alternative approaches such as discrete diffusion. However, masked diffusion, which has emerged as a popular choice due to its simplicity and effectiveness, reintroduces this inability to revise words. To overcome this, we generalize masked diffusion and derive the theoretical backbone of a family of general interpolating discrete diffusion (GIDD) processes offering greater flexibility in the design of the noising processes. Leveraging a novel diffusion ELBO, we achieve compute-matched state-of-the-art performance in diffusion language modeling. Exploiting GIDD's flexibility, we explore a hybrid approach combining masking and uniform noise, leading to improved sample quality and unlocking the ability for the model to correct its own mistakes, an area where autoregressive models notoriously have struggled. Our code and models are open-source: https://github.com/dvruette/gidd/
Operations Research (OR) is vital for decision-making in many industries. While recent OR methods have seen significant improvements in automation and efficiency through integrating Large Language Models (LLMs), they still struggle to produce meaningful explanations. This lack of clarity raises concerns about transparency and trustworthiness in OR applications. To address these challenges, we propose a comprehensive framework, Explainable Operations Research (EOR), emphasizing actionable and understandable explanations accompanying optimization. The core of EOR is the concept of Decision Information, which emerges from what-if analysis and focuses on evaluating the impact of complex constraints (or parameters) changes on decision-making. Specifically, we utilize bipartite graphs to quantify the changes in the OR model and adopt LLMs to improve the explanation capabilities. Additionally, we introduce the first industrial benchmark to rigorously evaluate the effectiveness of explanations and analyses in OR, establishing a new standard for transparency and clarity in the field.
Diffusion models have revolutionized generative tasks, especially in the domain of text-to-image synthesis; however, their iterative denoising process demands substantial computational resources. In this paper, we present a novel acceleration strategy that integrates token-level pruning with caching techniques to tackle this computational challenge. By employing noise relative magnitude, we identify significant token changes across denoising iterations. Additionally, we enhance token selection by incorporating spatial clustering and ensuring distributional balance. Our experiments demonstrate reveal a 50%-60% reduction in computational costs while preserving the performance of the model, thereby markedly increasing the efficiency of diffusion models. The code is available at https://github.com/ada-cheng/CAT-Pruning
Uncertainty estimation plays an important role for future reliable deployment of deep segmentation models in safety-critical scenarios such as medical applications. However, existing methods for uncertainty estimation have been limited by the lack of explicit guidance for calibrating the prediction risk and model confidence. In this work, we propose a novel fine-grained reward maximization (FGRM) framework, to address uncertainty estimation by directly utilizing an uncertainty metric related reward function with a reinforcement learning based model tuning algorithm. This would benefit the model uncertainty estimation through direct optimization guidance for model calibration. Specifically, our method designs a new uncertainty estimation reward function using the calibration metric, which is maximized to fine-tune an evidential learning pre-trained segmentation model for calibrating prediction risk. Importantly, we innovate an effective fine-grained parameter update scheme, which imposes fine-grained reward-weighting of each network parameter according to the parameter importance quantified by the fisher information matrix. To the best of our knowledge, this is the first work exploring reward optimization for model uncertainty estimation in safety-critical vision tasks. The effectiveness of our method is demonstrated on two large safety-critical surgical scene segmentation datasets under two different uncertainty estimation settings. With real-time one forward pass at inference, our method outperforms state-of-the-art methods by a clear margin on all the calibration metrics of uncertainty estimation, while maintaining a high task accuracy for the segmentation results. Code is available at \url{https://github.com/med-air/FGRM}.
Laparoscopic surgeries often suffer from reduced visual clarity due to the presence of surgical smoke originated by surgical instruments, which poses significant challenges for both surgeons and vision based computer-assisted technologies. In order to remove the surgical smoke, a novel U-Net deep learning with new loss function and integrated differentiable Wiener filter (ULW) method is presented. Specifically, the new loss function integrates the pixel, structural, and perceptual properties. Thus, the new loss function, which combines the structural similarity index measure loss, the perceptual loss, as well as the mean squared error loss, is able to enhance the quality and realism of the reconstructed images. Furthermore, the learnable Wiener filter is capable of effectively modelling the degradation process caused by the surgical smoke. The effectiveness of the proposed ULW method is evaluated using the publicly available paired laparoscopic smoke and smoke-free image dataset, which provides reliable benchmarking and quantitative comparisons. Experimental results show that the proposed ULW method excels in both visual clarity and metric-based evaluation. As a result, the proposed ULW method offers a promising solution for real-time enhancement of laparoscopic imagery. The code is available at https://github.com/chengyuyang-njit/ImageDesmoke.
Kolmogorov-Arnold Network (KAN) has attracted growing interest for its strong function approximation capability. In our previous work, KAN and its variants were explored in score regression for blind image quality assessment (BIQA). However, these models encounter challenges when processing high-dimensional features, leading to limited performance gains and increased computational cost. To address these issues, we propose TaylorKAN that leverages the Taylor expansions as learnable activation functions to enhance local approximation capability. To improve the computational efficiency, network depth reduction and feature dimensionality compression are integrated into the TaylorKAN-based score regression pipeline. On five databases (BID, CLIVE, KonIQ, SPAQ, and FLIVE) with authentic distortions, extensive experiments demonstrate that TaylorKAN consistently outperforms the other KAN-related models, indicating that the local approximation via Taylor expansions is more effective than global approximation using orthogonal functions. Its generalization capacity is validated through inter-database experiments. The findings highlight the potential of TaylorKAN as an efficient and robust model for high-dimensional score regression.
Balancing reconstruction quality versus model efficiency remains a critical challenge in lightweight single image super-resolution (SISR). Despite the prevalence of attention mechanisms in recent state-of-the-art SISR approaches that primarily emphasize or suppress feature maps, alternative architectural paradigms warrant further exploration. This paper introduces DiMoSR (Dilated Modulation Super-Resolution), a novel architecture that enhances feature representation through modulation to complement attention in lightweight SISR networks. The proposed approach leverages multi-branch dilated convolutions to capture rich contextual information over a wider receptive field while maintaining computational efficiency. Experimental results demonstrate that DiMoSR outperforms state-of-the-art lightweight methods across diverse benchmark datasets, achieving superior PSNR and SSIM metrics with comparable or reduced computational complexity. Through comprehensive ablation studies, this work not only validates the effectiveness of DiMoSR but also provides critical insights into the interplay between attention mechanisms and feature modulation to guide future research in efficient network design. The code and model weights to reproduce our results are available at: https://github.com/makinyilmaz/DiMoSR
A cardiac digital twin is a virtual replica of a patient's heart for screening, diagnosis, prognosis, risk assessment, and treatment planning of cardiovascular diseases. This requires an anatomically accurate patient-specific 3D structural representation of the heart, suitable for electro-mechanical simulations or study of disease mechanisms. However, generation of cardiac digital twins at scale is demanding and there are no public repositories of models across demographic groups. We describe an automatic open-source pipeline for creating patient-specific left and right ventricular meshes from cardiovascular magnetic resonance images, its application to a large cohort of ~55000 participants from UK Biobank, and the construction of the most comprehensive cohort of adult heart models to date, comprising 1423 representative meshes across sex (male, female), body mass index (range: 16 - 42 kg/m$^2$) and age (range: 49 - 80 years). Our code is available at https://github.com/cdttk/biv-volumetric-meshing/tree/plos2025 , and pre-trained networks, representative volumetric meshes with fibers and UVCs will be made available soon.