The paper presents a local traditional OCR pipeline that can be iteratively fine‑tuned on both layout and appearance levels of complex historical Sanskrit manuscripts. By adapting to the specific manuscript distribution, the pipeline improves transcription accuracy across subsequent pages, reducing the need for costly human annotation. The authors apply the method to three manuscripts, release a dataset with detailed layout and Unicode annotations in PAGE‑XML format, and benchmark the results against leading multimodal large language models.
DynCur-Geo introduces a dynamic curiosity framework for active geo‑localization, adjusting the intrinsic reward based on the remaining distance to a target. A distance‑aware gate promotes early exploration and transitions the policy toward goal‑directed behavior as the UAV approaches the target, while potential‑based reward shaping provides dense progress guidance. Experiments in multimodal, cross‑scene, disaster‑affected, and long‑range scenarios demonstrate consistent performance gains over existing active geo‑localization baselines.
Aligned vision‑language models (VLMs) are designed to combine grounded visual reasoning with safe generation. The study finds that when safety constraints are applied, these models often abstain from answering questions that they could answer under default instruction, yet visual evidence continues to influence the decoding process. The authors show that safety‑induced abstention alters late‑stage hidden‑state dynamics, and that targeted interventions can restore grounded answering without retraining or changing visual inputs.
The paper introduces Molmo2Fish, an interactive tool that uses a multimodal large language model to correct imperfect fish tracking predictions in sonar datasets. It demonstrates that the system can effectively improve multi‑object tracking performance through natural language guidance, though further enhancements are needed. The authors provide open‑source code and data for replication.
EXPO-FT is a system that enables stable, sample‑efficient reinforcement learning fine‑tuning of pretrained Vision‑Language‑Action (VLA) policies. It achieves perfect success on a range of manipulation tasks—such as routing string lights, striking a pool ball, and inserting a flower into a wine bottle—using only about 19.1 minutes of online robot data. The approach outperforms both RL-from-scratch and existing VLA fine‑tuning methods, and the authors provide an open‑source codebase to support wider adoption.
By Perry Dong, Kuo-Han Hung, Tian Gao, Dorsa Sadigh, Chelsea Finn
The 10th AI City Challenge, held alongside ECCV 2026, celebrates a decade of benchmarking for intelligent transportation, smart cities, and physical AI. Since its 2017 inception focused on vehicle detection, classification, and tracking, the challenge has expanded into a comprehensive benchmark suite covering multi‑camera perception, multimodal reasoning, synthetic‑to‑real learning, generative forecasting, and privacy‑preserving evaluation. The 2026 edition saw 325 registered teams from 26 countries, with six main tracks—spanning multi‑camera 3D perception, transportation safety captioning and VQA, traffic anomaly reasoning, text‑based person anomaly search, generative traffic video forecasting, and cross‑city object detection—plus two out‑of‑domain leaderboards for fisheye traffic‑violation understanding and pedestrian situated‑intent VQA.
By Zheng Tang, Shuo Wang, David C. Anastasiu, Ming-Ching Chang, Anuj Sharma, Quan Kong, Munkhjargal Gochoo, Jun-Wei Hsieh, Tomasz Kornuta, Zhedong Zheng, Renran Tian, Judah Goldfeder, Fulgencio Navarro, Yuxing Wang, Yizhou Wang, Sameer Satish Pusegaonkar, Anqi Li, Nalin Dadhich, Ridham Kachhadiya, Dhanishtha Patil, Haoquan Liang, Jiajun Li, Han Zhang, Yilin Zhao, Zaid Pervaiz Bhat, Shuyu Yang, Ashutosh Kumar, Rong Wang, Rafael Martin Nieto, Peter Christiansen, Ahmed Abduljawad, Mohanrasu Shanmugam, Nadeem Shaik, Sujit Biswas, Xunlei Wu, Vidya Murali, Rama Chellappa
The paper introduces a novel framework that converts structural framing plans from PDF drawings into editable finite‑element model drafts without requiring task‑specific detector training. It combines a deterministic geometry extraction stage—estimating scale, recognizing five entity classes, and assembling a drafting grammar—with an agentic vision‑language refinement stage that proposes corrections, performs admission tests, and ensures fail‑closed transactions. Evaluation on a 100‑plan benchmark shows high accuracy, with scale within 0.1% and recall/precision values above 0.86 for all component types.
By Mohammad Talebi-Kalaleh, Qipei Mei
The paper introduces MemJack, a memory‑augmented multi‑agent framework that automatically generates jailbreak attacks on Vision‑Language Models (VLMs) using benign natural images as visual anchors. MemJack discovers visual anchors, camouflages them semantically, evaluates responses, repairs via reflection, and replans dynamically, forming a closed‑loop attack pipeline. The authors also create MemJack‑Bench, a dataset of over 113,000 interactive multimodal jailbreak trajectories, and show that MemJack achieves a 71.48% attack success rate against Qwen3‑VL‑Plus, reaching 90% under extended budgets, outperforming other baselines on natural‑image evaluation.
By Jianhao Chen, Haoyang Chen, Hanjie Zhao, Haozhe Liang, Zheng Wang, Tieyun Qian
MultiSigBERT is a unified framework that performs multimodal sequential survival modeling in oncology by integrating narrative clinical reports, numerical measurements, and structured variables. The method converts free-text reports into sentence embeddings, compresses them with modality-specific PCA, and concatenates them with structured covariates to create joint temporal trajectories. These trajectories are encoded using the Signature transform from Rough Paths theory, and the resulting high-dimensional features are fed into a LASSO-regularized Cox model, achieving a concordance index of 0.743 on an independent test set of over 2,500 patients.
By Paul Minchella, St\'ephane Chr\'etien, Guillaume Metzler, Lo\"ic Verlingue, R\'emi Vaucher
SCENARIODIFF is a hierarchical contextual reasoning framework designed for multimodal time series forecasting, especially in event-driven domains. It processes textual context through three agents—Historical Context, Scenario, and Anchor Guidance—to generate structured signals that condition a Multimodal Diffusion Transformer. The framework also employs Anchor Blended Sampling to locally refine forecast trajectories without retraining, and demonstrates superior performance on the Time‑MMD benchmark.
By Tuan-Binh Tran, Dat Nguyen Cong, Duc-Trong Le, Thanh Trung Huynh, Tung Kieu
VLCP (Vision Language Control Policy) is a training‑free robot manipulation approach that keeps a vision‑language model (VLM) frozen and uses it to generate short Python control functions. Unlike traditional methods that retry a fixed policy, VLCP rewrites the control code every K steps based on multi‑view RGB, proprioceptive state, and state delta, allowing failures to be corrected within the same episode. In a 57‑task MuJoCo/RoboVerse benchmark, VLCP achieves 35.1% pooled success versus 3.5% for a single‑query baseline, with a 27.3% within‑episode recovery rate on failed grasps and efficient token usage.
By Dhia Naouali, Minghan Wu, Claudia Wong, Abhinav Puthran, Omar G. Younis
Prism‑GRPO enhances the GRPO reinforcement‑learning algorithm for vision‑language‑action policies by adding a weighted trajectory‑level execution‑quality score to binary success rewards. This approach splits groups with identical outcomes into a quality spectrum, preserving training signal while ensuring successes always outrank failures. Experiments on four RoboTwin tasks show Prism‑GRPO achieves higher success and quality at matched rollout budgets, reaching target success rates with up to 56% fewer rollouts and mitigating reward‑hacking behaviors that transfer to real‑robot deployment.
By Zeyun Deng, Yuzhe Lu, Yawei Wang, Linbo Liu, Qing Ping, Han Ding, Guande Wu, Panpan Xu, Jun Huan
The paper introduces TiMi, a framework that enhances time series transformers with a Multimodal Mixture-of-Experts (MMoE) module to incorporate multimodal data, especially textual information, into forecasting. TiMi leverages large language models to generate future inferences that guide predictions, eliminating the need for explicit representation alignment. Experiments show TiMi achieves state‑of‑the‑art performance on sixteen real‑world multimodal forecasting benchmarks, outperforming advanced baselines while maintaining adaptability and interpretability.
By Jiafeng Lin, Yuxuan Wang, Huakun Luo, Jianmin Wang, Zhongyi Pei
PathPocket is a multimodal AI co‑pilot that grounds pathology decision‑making in evidence. It builds the largest pathology evidence corpus (≈110,472 documents) and a hypergraph of 4.55 million entities and 7.10 million relations to support traceable reasoning. The system handles text and multimodal queries, including ROI and gigapixel whole‑slide images, and outperforms current state‑of‑the‑art models on a benchmark of over 200,000 real‑world cases, improving pathologists’ diagnostic accuracy and confidence.
By Zhe Xu, Zhengyu Zhang, Zhiyuan Cai, Jiahao Xu, Yijie Lin, Ziyi Liu, Junlin Hou, Hongyi Wang, Yuxiang Nie, Yihui Wang, Jiabo Ma, Ling Liang, Yingxue Xu, Zhengrui Guo, Guanghao Wu, Danyi Li, Ziqi Zhou, Donglin Tan, Zhijian Cen, Ying Tan, Xiaolin Liu, Qi Xie, Xiaoying Tang, Xi Peng, Cheng Deng, Lijuan Qu, Ronald Cheong Kin Chan, Li Liang, Hao Chen
The paper introduces a new strategy for connecting large language models (LLMs) to speech encoders in automatic speech recognition (ASR) systems by sharing a single connector across languages within the same linguistic family. This approach reduces the number of parameters needed compared to training a separate connector for each language, while improving generalization across different domains and real‑world corpora. Experiments with two multilingual LLMs and two speech datasets demonstrate that family‑based connectors are both efficient and effective for multilingual ASR deployment.
By Yuchen Zhang, Ravi Shekhar, Haralambos Mouratidis
The paper investigates the expressive power of multimodal contrastive learning architectures by treating them as parameterized families of joint density estimators. It shows that the classic two‑tower CLIP model is a universal approximator for two modalities, while a common extension that sums pairwise similarities fails to approximate arbitrary joint distributions when three or more modalities are involved, though it can match all pairwise conditionals. To address this limitation, the authors introduce Hadamard‑CLIP, which adds a single learned weight vector to restore universal approximation for any number of modalities while retaining CLIP’s efficient retrieval capabilities.
By Andrew Stuart, Florian Wolf
The paper introduces COMIC, a reference‑aware safety gate designed for multimodal large language models (MLLMs). COMIC detects the operation requested by a user, identifies visual targets through OCR and open‑vocabulary proposals, and evaluates safety on explicit operation‑target pairs, using max‑risk aggregation and quality‑aware routing to decide whether to allow or block a request. Experiments on several open‑source MLLMs and jailbreak benchmarks show that COMIC improves robustness while maintaining benign utility and efficiency.
By Md Abdullahil Oaphy, Anhao Xiang, Zongxing Xie, Huayue Gu, Chenyu Wang, Honghui Xu
Co‑RL is a multi‑agent reinforcement learning framework that trains several decoupled models without shared parameters, using rewards generated by their peers. By increasing cohort diversity—through heterogeneous model families, varying sizes, and rephrased training samples—Co‑RL reduces self‑reinforcing feedback loops, preserves behavioral diversity, and prevents training collapse. Across both text‑only and multimodal benchmarks, Co‑RL outperforms base models and prior label‑free methods, achieving gains of 3.0‑8.6% on seven text benchmarks and 2.3‑7.2% on four multimodal benchmarks, while matching or surpassing supervised approaches without any ground‑truth labels.
By Yunhao Yang, Yuexin Bian, Yunjie Tian, Di Fu, Tianjin Huang, Yuanyuan Shi, Ziang Xiao, Nuno Vasconcelos, Yijiang Li
DeAR (Decentralized Agentic Reasoning) is a new framework that replaces centralized protocols with autonomous peer‑to‑peer collaboration for agentic reasoning. It introduces three key mechanisms: decentralized capability grounding for agent specialization, thought map navigation for targeted peer interactions, and topology updates for adaptive error correction. Across nine multimodal reasoning and text‑based QA benchmarks, DeAR consistently outperforms recent baseline methods, demonstrating that decentralized and adaptive collaboration improves accuracy in knowledge‑intensive reasoning tasks.
By Xing Wei, Changmeng Zheng, XiaoYong Wei, Xiufen Ye, Qing Li
CORAM (Coherent Orthogonal Rotation for Model Merging) is a new method for combining fine‑tuned models without joint training or access to original data. It partitions each target weight matrix into row slices, represents each expert slice with its singular value decomposition in the base‑model SVD frame, and merges the task‑specific factors on their corresponding manifolds. The approach includes an amplification coefficient to counteract manifold averaging contraction, spread slicing to balance highly updated rows, and a residual pathway for non‑target layers, achieving improvements over existing orthogonal merging techniques across multiple model families and scales.
By Xinyi Sui, Ziran Liu, Nam Ling, Wei Wang, Wei Jiang