The paper identifies a flaw in current text-to-video safety filters that rely on keyword prompts and per-frame checks, showing that these systems miss harms that arise solely from the ordering of frames. Experiments reveal that a significant portion of generated videos contain harmful content that is undetected because each individual frame appears safe, and prompt engineering does not mitigate this issue. The authors propose an order-aware detector that can distinguish harmful videos from benign shuffles, achieving an AUC of 0.74 compared to chance for per-frame checking.
By Yuxin Cao, Fusen Guo, Yuezhong Wu, Huadong Mo, Wei Song
The paper argues that token importance alone is insufficient to determine safe removal of visual tokens in multimodal large language models, because removability depends on representation depth and the surrounding deletion set. Through controlled experiments, the authors show that the same tokens can have different effects when removed at different depths or contexts. They introduce CoRePrune, a training‑free two‑stage pruning framework that refreshes deletion effects as visual representations evolve and refines candidate tokens based on the current deletion set, achieving high performance retention across multiple backbones and reducing prefill time significantly.
By Shengli He, Yongchao Liang, Roumeng He, Junjie Zeng, Jiyuan He, Xin Fang, Can Wu, Li Zheng
Semantically-Guided Domain Randomization (S‑GDR) is an annotation‑free pipeline that uses vision‑language model captioning of a small real reference set, diffusion‑based background synthesis, and mask‑based object composition to generate synthetic training data. In a high‑mix, low‑volume automotive detection benchmark, S‑GDR achieves a mAP50‑95 of 0.739 with only 200 synthetic images, outperforming a domain‑randomized render baseline and several other synthetic data methods under the same budget. These results suggest S‑GDR is a viable alternative for training visual perception systems when annotation, energy, and time resources are severely limited.
By Jose Moises Araya-Martinez, Gautham Mohan, Jens Lambrecht
The paper investigates how multimodal language models can generate perceptual representations without dedicated encoders. It shows that the shared transformer can internally create these representations in its early-to-middle layers, a structure termed a Virtual Encoder. Experiments with linear probing, similarity metrics, and causal analysis reveal that this encoder-like computation emerges even when models receive only perceptual tokens, indicating that perception and language processing can be decoupled within a single architecture.
By Katsuya Ogata, Yuta Nakashima
HARMONY is a hierarchical chain-of-thought framework that reconstructs complete 3D indoor scenes from a single monocular image. It combines agentic reasoning with visual geometry foundation models, starting with camera calibration and semantic layout recovery, then placing objects hierarchically while refining geometry with point cloud estimations. The method achieves semantically consistent scenes that align perceptually with the input image, outperforming existing baselines on synthetic and real-world data.
By Shufan Sun, Chen Wang, Enxin Song, Jiatao Gu, Lingjie Liu
The paper introduces NPLSD, a pair of line‑segment detectors optimized for the Neural‑ART NPU on STM32N6 microcontrollers. NPLSD‑H retains a convolutional backbone and replaces transformer components with a fully‑convolutional head, while NPLSD‑M adapts a lightweight trunk to the NPU’s supported operators. Trained on ImageNet and ShanghaiTech Wireframe, the models achieve competitive accuracy with 2.63 M and 0.62 M parameters, respectively, and ablation studies show initialization contributes significantly to performance.
By Parsa Hassani Shariat Panahi, Amir Hossein Jalilvand, M. Hassan Najafi
arXiv:2609.25040v1 Announce Type: cross
Abstract: Banana crop diseases threaten food security across the world, yet field diagnosis remains difficult because of limited expert access and visual simil...
By Sangam Kumar Jena, Pandarasamy Arjunan
arXiv:2505.01583v2 Announce Type: replace
Abstract: Understanding causal event relationships and achieving fine-grained temporal grounding in videos remain challenging for vision-language models (VLM...
By Jen-Hao Cheng, Yi-Hao Peng, Huapeng Zhou, Vivian Wang, Huayu Wang, Hsiang-Wei Huang, Wenhao Chai, Hou-I Liu, Kuang-Ming Chen, Cheng-Yen Yang, Yi-Ling Chen, Vibhav Vineet, Qin Cai, Jenq-Neng Hwang
arXiv:2602.01163v2 Announce Type: replace
Abstract: Safe UAV emergency landing requires more than just identifying flat terrain; it demands understanding complex semantic risks (e.g., crowds, tempora...
By Chunliang Hua, Lei Zhang, Jiayang Sun, Chunlan Zeng, Xiao Hu
arXiv:2603.17043v2 Announce Type: replace
Abstract: Physics-aware multimodal large language models (MLLMs) can localize and characterize two-dimensional (2D) material flakes from optical microscopy i...
By Sankalp Pandey, Thanh-Dat Truong, Xuan-Bac Nguyen, Hoang-Quan Nguyen, Tim Faltermeier, Nicholas Borys, Hugh Churchill, Khoa Luu
arXiv:2606.22694v2 Announce Type: replace
Abstract: Vision-Language Models (VLMs) remain unreliable when spatial reasoning requires composing relations whose meanings depend on frames of reference. E...
By Danial Kamali, Tanawan Premsri, Shreya Rajpal, Amir Zadeh, Chuan Li, Parisa Kordjamshidi
arXiv:2503.06989v5 Announce Type: replace-cross
Abstract: Recently, Multimodal Large Language Models (MLLMs) have demonstrated their superior ability in understanding multimodal content. However, the...
By Wenzhuo Xu, Zhipeng Wei, Xiongtao Sun, Zonghao Ying, Deyue Zhang, Dongdong Yang, Xiangzheng Zhang, Quanchen Zou
arXiv:2606.22892v2 Announce Type: replace-cross
Abstract: The clinical diagnosis of skin diseases is susceptible to interference from inter-class similarity of skin lesions, and over-reliance on clin...
By Haibiao Li, Di Lin, Xue Jiang, Weiwei Wu, Yanxi Li, Yugang Chi
MultiViewDx is a physician‑validated multimodal instruction dataset that links medical imaging studies with patient context and normalizes heterogeneous reports into an evidence‑linked workflow (evidence → findings → differential discussion → diagnosis). The dataset covers a wide range of imaging modalities and uses a unified image‑text retriever to ensure that instruction synthesis is grounded in source‑supported evidence. Fine‑tuned models on MultiViewDx achieve the highest average accuracy on four MedVQA benchmarks and receive the strongest overall rating on JAMA Clinical Challenge cases, with ablations confirming the importance of case‑level multi‑view organization and evidence‑linked reasoning.
By Junda Wang, Zonghai Yao, Yujan Ting, Eric Z. Chen, Hieu Tran, Hong Yu, Weijing Huang, Terrence Chen
arXiv:2609.25447v1 Announce Type: new
Abstract: We study what LLMs do when a user applies pressure in an uncomfortable situation: a user insists, begs, flatters or grieves, and the model gives up a c...
By Tapan Parikh
arXiv:2609.22497v1 Announce Type: new
Abstract: The aggregation of many lay estimates often outperforms individual expert judgment, a phenomenon known as the wisdom of crowds. While this is usually a...
By Federico Barrera-Lemarchand, Mariano Sigman, Joaquin Navajas
arXiv:2609.22512v1 Announce Type: new
Abstract: Consensus among LLM judges is often taken as strong evidence that a decision is correct. This assumes that judges make their errors independently. In p...
By Elias Hossain, Niloofar Yousefi, Ser-Nam Lim
arXiv:2609.22537v1 Announce Type: new
Abstract: Enterprise AI assistants must produce responses that are verifiable and traceable to source evidence. However, retrieval augmented generation (RAG) ove...
By Anubha Kabra, Katie Jooyoung Kim, Colin Zhiwei Kou, Helene Sajer, Yimei Fan, Radomir Cisar, Heather Greenhalgh, Gabriel Martinez Vidiri
arXiv:2609.22696v1 Announce Type: new
Abstract: Decentralized social media platforms create new opportunities and challenges for computational mental health research because data access, moderation,...
By Gaurab Chhetri, Anandi Dutta, Subasish Das
arXiv:2609.22746v1 Announce Type: new
Abstract: Large Language Models (LLMs) have recently been introduced into traffic signal control (TSC) as decision agents due to their strengths in human-readabl...
By Huaitao Zhao, Tianlong Zhou, Weijie Wang, Jiasheng Shi, Weixiong Rao