OmniHallu is a unified framework for detecting hallucinations in multimodal large language models across both comprehension and generation tasks involving image, video, and audio modalities. It introduces OmniHallu-Bench, a 10,000-sample benchmark with claim-level human annotations for six cross-modal tasks (I2T, V2T, A2T, T2I, T2V, T2A). The system uses a multi‑agent architecture that decomposes outputs into atomic claims, verifies them with modality‑specific experts, and aggregates evidence through structured reasoning, while a preference‑optimized verifier reduces expert calls by 66% with minimal performance loss.
By Jianjiang Yang, Peihang Li, Shanqing Xu, Mengchen Qian, Lu Zhang, Meng Luo
The paper presents the first controlled study of Whisper adaptation for Greek Automatic Lyric Transcription (ALT), addressing challenges such as melodic variability, rhythmic irregularity, and accompaniment interference. It explores model scaling, multitask training with transcribe-translate ratios, and a two‑stage speech‑to‑singing adaptation, while curating a segment‑level aligned singing dataset from the Greek Audio Dataset. Results show that larger models consistently improve performance, multitask learning benefits smaller models, and the two‑stage adaptation achieves a 27.2% WER, establishing the first Greek ALT benchmark.
By Maria Frangiadaki, Dimitrios Damianos, Kosmas Kritsis, Vassilis Katsouros
SWRouter is a new routing method for multi‑turn large language model conversations that uses a similarity‑based context segmentation mechanism to construct prompts and a dual‑metric evaluation framework to separate construction accuracy from router performance. The approach addresses two key challenges in multi‑turn dialogue: preventing information loss or confusion during context construction and evaluating routing quality independently of prompt quality. Experiments on multi‑turn dialogue benchmarks show that SWRouter outperforms strong baselines, improving evaluation accuracy by 16.26% over the best individual large language model and by 8.22% over the Conv‑ID Context baseline.
By Yu Wang, Yuchen Li, Rui Kong, Xinran Chen, Jiamin Chen, Hengyi Cai, Shuaiqiang Wang, Jiashu Zhao, Yulun Zhang, Zhonghao Lyu, Haoyi Xiong, Linghe Kong, Jimmy Xiangji Huang, Dawei Yin
IndicTriMix presents a new benchmark and models for token‑level language identification in tri‑language code‑mixed text involving Hindi, Gujarati, and Bengali. The authors reformulate the task as sequence labeling and fine‑tune transformer models MuRIL and XLM‑RoBERTa, evaluating them on manually annotated test sets. They also introduce two code‑mixing generation methods using parallel sentences and release the datasets and fine‑tuned models for public use.
By Pruthwik Mishra, Rudra Trivedi, Avi Patel, Ashok Urlana, Shrikant Malviya
The paper presents the KuaiRP series of role‑playing models, detailing a multi‑stage training pipeline that balances deep domain knowledge injection with the preservation of general agent capabilities. The approach includes a standardized character template, a supervised fine‑tuning (SFT) data pipeline, a rule‑based reward function for reinforcement learning, and a novel two‑stage on‑policy distillation (OPD) with cumulative‑divergence decay (CDD) to recover general skills. Experimental results show that the models achieve state‑of‑the‑art role‑playing fidelity in target domains while maintaining low deployment costs.
By Yipeng Wang, Ziwei Zhang, Jiahui Zhang, Qi Gan, Kai Sheng
RetroThinker is a multi-stage post‑training framework that enhances SpeechLLMs by enabling them to self‑verify and forward‑correct Chain‑of‑Thought reasoning steps during inference. It combines supervised fine‑tuning on curated retrospective thinking data with length‑based direct preference optimization to improve reasoning while the user speaks. On the GSM8K benchmark, RetroThinker achieves an 11% absolute accuracy gain over non‑retrospective baselines while maintaining comparable latency.
By Yi-Jen Shih, Puyuan Peng, Abdelrahman Mohamed, David Harwath
Deep Research Bench II is a new benchmark designed to evaluate Deep Research Agents (DRAs) by requiring them to produce research reports for 132 grounded tasks across 22 domains. Each report is assessed using 9,430 fine‑grained binary rubrics that cover information recall, analysis, and presentation, all derived from expert‑written investigative articles through a rigorous LLM‑plus‑human pipeline. Evaluation of current state‑of‑the‑art DRAs shows that even the best models satisfy fewer than 50% of these rubrics, highlighting a significant gap between automated agents and human experts.
By Ruizhe Li, Mingxuan Du, Benfeng Xu, Chiwei Zhu, Xiaorui Wang, Zhendong Mao
The paper introduces CoCoEval, a framework for evaluating large language model (LLM)–simulated conversations by detecting 10 types of inconsistent and uncollaborative behaviors at the turn level. Using CoCoEval, the authors compare human conversations with those generated by GPT‑4.1, GPT‑5.1, and Claude Opus 4, finding that LLMs produce far fewer such behaviors under vanilla prompting and that prompt engineering or fine‑tuning often over‑produces specific behaviors. The study highlights gaps between human and LLM‑simulated interactions that conventional Likert‑scale evaluations miss, raising concerns about using LLMs as proxies for human social interaction.
By Ryo Kamoi, Ameya Godbole, Binglin Zhou, Xiaoxin Lu, Longqi Yang, Rui Zhang, Mengting Wan, Pei Zhou
The paper introduces MERIT, a training‑free agent that uses an online dual‑polarity memory of oracle‑verified corrections and failed attempts to improve LLM agent repair without updating parameters. MERIT retrieves only finalized memories, classifies failures, and employs a hybrid lexical‑dense retriever before the frozen model generates revisions. Experiments with Qwen2.5‑7B‑Instruct show that MERIT raises execution accuracy on the Spider benchmark from 66.34 % to 69.79 % and on BIRD from 47.35 % to 48.44 %, with varying evidence across datasets.
By Khang Nhat Hoang Vo, Tam Minh Chu, Anh Trac Duc Dinh, Thuyen Vinh Ha Bui, Tho Quan
The paper introduces a new framework for learning continuous-time diffeomorphic image registration by modeling a non-autonomous ODE as a two-parameter flow map. By enforcing cocycle consistency, the method learns flow maps without time discretization or velocity integration during training, enabling efficient inference with few compositions. Experiments on nine datasets show consistent alignment improvements, including a 2.1% Dice gain on brain MRI, 12% TRE reduction on lung CT, and 2.6% Dice improvement on cardiac MRI and ultrasound.
By Mohammadjavad Matinkia, Nilanjan Ray
HiPerViT is a compact vision-only architecture that injects an explicit second-order statistical prior into a transformer-based pipeline for texture recognition. It combines global and local image views with a compact bilinear descriptor encoded as a statistical token, and integrates this token with first-order spatial representations through Perceiver-style latent distillation. Across six texture recognition benchmarks, HiPerViT consistently outperforms strong vision-only baselines, achieving notable gains on DTD, GTOS-Mobile, and 1200Tex, and the improvements are largely independent of backbone depth or fusion topology.
By Jo\~ao Pedro C. A. de S\'a, Odemir Martinez Bruno
The paper presents M2Heat, a physics-inspired framework for fusing hyperspectral and LiDAR data in land‑cover classification. It introduces a visual heat conduction module (vHeat) and Frequency Value Embeddings (FVEs) to model anisotropic information flow, capturing global dependencies with sub‑quadratic complexity. Combined with a Cross‑Frequency Fusion (CFF) module, M2Heat delivers discriminative, robust features and achieves competitive performance on Trento, Houston2013, and Augsburg benchmarks while offering an interpretable heat‑conduction perspective.
By Kan Wei, Jiahui Cui, Jing Yao, Xinyu Zhao, Lei Wang, Pedram Ghamisi
ReconPlusGen introduces a method that injects a reconstruction prior into multi‑view 3D generation. By predicting a point cloud in canonical space from multiple input images, the method deterministically injects the geometry into a diffusion process via noise inversion and then modulates the noise to maintain generative flexibility for completing unseen areas and refining visible geometry. The paper presents qualitative results on benchmark and real‑world images, along with an illustration of the reconstruction‑guided noise initialization and modulation.
By Jiarui Liu, Heng Li, Weiyu Li, Keng Deng, Junyuan Deng, Zheng Zhongxing, Junyu Huang, Jiahao Chang, Xiaoguang Han, Ping Tan
The paper introduces a multi‑view, confusion‑guided ensemble framework for synthetic image attribution, combining FFT‑ConvNeXt, DINOv2, CLIP, and Xception to capture frequency, semantic, and forensic cues. Extensive data augmentation simulates realistic post‑processing, while a binary expert classifier and class‑adaptive confidence calibration address ambiguities between similar diffusion models. The approach achieved 99.53% on the public leaderboard and 99.20% on the private leaderboard for the ICANN 2026 DLMMDD Workshop challenge.
By Zuomin Qu
arXiv:2609. 11235v1 Announce Type: new Abstract: Test-time adaptation (TTA) offers many ways to update a deployed model without labels, but choosing the wrong update can make a strong source model worse.
By Kartik Jhawar, Lipo Wang
The paper introduces VWG-Bench, a benchmark covering nine reasoning dimensions and 38 tasks to evaluate video generative models on symbolic reasoning, physical laws, and goal pursuit. It also presents Vid-PRE, a prompt-rewriting framework that offloads reasoning to a VLM, improving logical performance without changing the generator architecture. Experiments show that current models excel at visual quality but struggle with logic-heavy tasks, while Vid-PRE significantly boosts reasoning across multiple generators.
By Meng Luo, Yicheng Liu, Jiahao Wang, Yuanxing Zhang, Xin Tao, Pengfei Wan, Kun Gai, Hao Fei
The paper introduces MOCO, a diffusion-based framework that generates 3D avatar motions from concurrent multimodal inputs such as speech audio, text descriptions, and trajectory data. MOCO decouples motion generation by independently producing modality-specific motions at each denoising step and then assembling them according to spatial rules, iteratively refining the combined motion. This approach yields coherent, lifelike, and synchronized movements, outperforming existing baselines on a multimodal benchmark.
By Yifei Liu, Qiong Cao, Hongwei Yi, Huaiguang Jiang, Changxing Ding
The paper presents a segmentation pipeline for brain metastases in both pre‑ and post‑treatment cases using a 5‑fold nnU‑Net ResEnc‑L ensemble trained on 1,296 four‑modality cases. A rule‑based post‑processing cascade improves the lesion‑wise Dice similarity coefficient (LW‑DSC) for enhancing tumour, tumour core, whole tumour, and resection cavity sub‑regions, achieving LW‑DSC scores of 0.733, 0.751, 0.713, and 0.549 respectively on the official validation leaderboard. The authors conduct a five‑fold out‑of‑fold analysis to validate the robustness of each post‑processing stage, provide a mechanistic explanation of LW‑DSC behaviour, and report thirteen negative results that challenge common intuitions, with all code released under Apache‑2.0.
By Haobin Liu, Xin Wang
OmniKVQuant introduces a training‑free framework for quantizing the key‑value (KV) cache of omni‑modal large language models (Omni‑LLMs) that process audio, video, and text simultaneously. The method addresses two identified problems—temporal key drift and heterogeneous value geometry—by setting key quantization ranges over short input windows and rotating values separately for each modality. Applied to Qwen2.5‑Omni and Qwen3‑Omni, OmniKVQuant achieves 2‑bit KV caches while largely preserving performance across seven audio‑visual benchmarks, and includes a fused Triton decode kernel that eliminates the need for a dense FP16 cache.
By Suho Yoo, Hyunjong Ok, Jongmin Choi, Jihoo Jung, Joon Son Chung
LangStreet introduces a persistent language field for anchor‑decoded street Gaussian scenes, addressing the breakdown of language Gaussian fields when geometry and appearance vary with camera views. The method relies on semantic ownership, where persistent decoder slots and parent anchors own the language field while transient children route observations. By accumulating additive directional evidence at slots and completing weakly supported slots with anchor‑aligned evidence, LangStreet achieves near‑maximal performance with significantly reduced storage, as demonstrated on KITTI, Virtual KITTI, and Waymo datasets.
By Runyi Yang, Deheng Zhang, Xiaoye Wang, Mengjiao Ma, Lei Sun, Kanzhi Wu, Ajad Chhatkuli, Luc Van Gool, Danda Pani Paudel