Editable Visual Design introduces a new design paradigm that combines a Coding Agent with a Vision‑Language Model (VLM) and an image generation model. The VLM acts as the creative brain, understanding requirements, planning tasks, and judging aesthetics, while the image generator produces isolated visual assets on demand. The agent follows an "imagine first, then act" workflow, generating assets, writing native HTML/CSS, and refining the design through visual feedback, ultimately producing editable, layer‑wise artifacts with real text that can be adjusted via a graphical interface.
By Junyan Ye, Wei Liu, Dongzhi Jiang, Zichen Wen, HaoDong Li, Zhutao Lv, Jiaxin Lin, Jinhua Yu, Jun He, Zilong Huang, Rui Chen, Weijia Li
The paper introduces a reverse sign‑language dictionary that recognizes signs from continuous signing without relying on gloss labels. It does this by first captioning a sign‑level video clip into a free‑form procedural description using an open‑weight vision‑language model, then retrieving the closest entry from a multilingual sentence encoder’s vocabulary of target descriptions. Experiments on a Japanese Sign Language dialogue corpus show that fine‑tuning the captioner boosts seen‑class retrieval from 4.5 % to 49 % and improves unseen‑class retrieval from 11.5 % to 21 %, approaching the performance of a standard closed‑set classifier while enabling open‑vocabulary recognition.
By Santiago Poveda-Guti\'errez, Hideki Nakayama, Mayumi Bono
The paper investigates latent visual reasoning in multimodal large language models, treating input, latent tokens, and final answer as a causal chain. Causal mediation analysis reveals two disconnections: latent tokens largely ignore input perturbations, and changes to latent tokens minimally affect the final answer, indicating limited causal influence. Probing shows latent tokens encode little visual information and are highly similar, leading the authors to propose CapImagine, an explicit text‑based imagination approach that outperforms latent‑space baselines on vision‑centric benchmarks.
By You Li, Chi Chen, Yanghao Li, Fanhu Zeng, Kaiyu Huang, Jinan Xu, Maosong Sun
The paper introduces EASE, a method that enhances multimodal reinforcement learning with verifiable rewards (RLVR) by adding visual‑evidence process supervision. EASE transforms annotated evidence regions into smoothed visual‑token targets and uses them to guide attention during RL training, but only on high‑reward trajectories. Experiments on Qwen2.5‑VL‑7B, Qwen3‑VL‑4B, and Qwen3‑VL‑8B show that EASE improves average scores over DAPO by 2.5 to 3.1 points across perception, hallucination, visual math, and multimodal reasoning benchmarks, and diagnostics confirm better alignment of visual attention with annotated evidence.
By Ruina Hu, Chen Wang, Lai Wei, Jionghao Bai, Bin Yu, Weiran Huang, Kai Wang, Yue Wang
The paper investigates when frozen vision‑language models (VLMs) respond to edits made at the representation level, specifically by modifying object‑token sets instead of the raw image. It introduces an answer‑key‑free protocol that evaluates edits without annotated post‑edit answers, revealing that VLM responses depend on explicit edit teaching, token cleanliness, density, and a separable reading axis. The study demonstrates that image‑free token edits can preserve most of the performance of matched patch‑token baselines and even outperform oracle methods on the VRSBench dataset across multiple remote‑sensing datasets and model backbones.
By Wonbin Son, Gyumun Choi, Junil Seo, Seungmin Rho, Mi Young Lee, Hyungjoon Kim
CoFiE introduces a two‑stage evidence selection framework for streaming video understanding, separating a coarse, query‑agnostic filtering of visually distinctive frames from a fine, query‑specific refinement during LLM prefill. By filtering out redundant frames before expensive vision encoding, CoFiE reduces end‑to‑end latency while maintaining high accuracy. The method achieves state‑of‑the‑art performance on benchmarks such as StreamingBench and OvO‑Bench, improving accuracy by up to 3.15% and inference speed by up to 2.54× compared to prior approaches.
By Jing Jiang, Yiran Ling, Ruonan Li, Dimitrios Stamoulis, Jie Liu
The paper introduces FactoSR, a factorized reinforcement learning framework designed to improve spatial reasoning in Vision‑Language Models by addressing a dimensional mismatch between 2D visual inputs and the 3D+temporal nature of the physical world. FactoSR decomposes the reasoning task into three orthogonal geometric sub‑objectives—planar correspondence (XY), depth consistency (Z), and temporal reversibility (T)—and optimizes these constraints within a unified policy learning mechanism. Experiments on multi‑view and video benchmarks show that this decomposition yields significant performance gains, achieving a 5.9% improvement on VSI‑Bench and 4.5% on All‑Angles‑Bench.
By Yijun Yang, Shenghe Zheng, Wenbo Li, Jianhui Liu, Haoze Sun, Yanbing Zhang, Jiaxiu Jiang, Lin Song, Haoyang Huang, Nan Duan, Lei Zhu
ProAct is a dual‑system framework for real‑time embodied social interaction that separates a low‑latency Behavioral System, which streams multimodal interaction and generates continuous non‑verbal motion, from a slower Cognitive System that performs long‑horizon social reasoning and produces proactive intentions. The Cognitive System uses an efficient memory mechanism and a user‑motivation prediction module to decide when to intervene, while the Behavioral System translates these intentions into fluid motion via an intention‑conditioned streaming flow‑matching generator with a disentangled ControlNet branch. The framework is deployed on a physical humanoid robot and validated through real‑world user studies, motion‑generation benchmarks, and a new ProActBench benchmark for proactive trigger detection and restraint.
By Zeyi Zhang, Zixi Kang, Ruijie Zhao, Yusen Feng, Biao Jiang, Hanyu Ji, Libin Liu
The paper presents a method for emotion recognition in virtual reality where head‑mounted displays occlude the upper face. By fusing lower‑face video with electromyography (EMG) signals from the occluded upper face, the authors achieve a 51% macro‑F1 score across seven emotional categories, outperforming image‑only and EMG‑only baselines. A new synchronized multimodal dataset from 20 participants is introduced and will be shared under an ethical‑use agreement.
By Birgit Nierula, Karam Tomotaki-Dawoud, Mert Akguel, Mustafa Tevfik Lafci, David Przewozny, Anna Hilsmann, Peter Eisert, Sebastian Bosse
Drive‑HWM introduces a hierarchical slow‑fast world modeling framework for autonomous driving. The slow model predicts multi‑step future representations, while the fast model jointly predicts the next frame and immediate action using a lightweight multimodal backbone and an autoregressive expert. Dynamic‑Aware Latents, learned through optical‑flow prediction, explicitly capture motion dynamics, and experiments on NAVSIM v1 and v2 show strong driving performance with validated ablation studies.
By Zhaoxin Fan, Tianbao Zhang, Wenjun Wu, Xiaofeng Wang, Yeying Jin, Jian Zhao, Zheng Zhu, Shuicheng Yan
The paper argues that BLEU-4, the prevailing metric for sign language translation (SLT), may not accurately reflect sign language proficiency because SLT models can exploit spurious correlations and spoken-language priors. By evaluating six SLT models on Phoenix-2014T and CSL-Daily, the authors show that higher BLEU-4 scores do not necessarily indicate better spatio-temporal understanding. They propose a new open-weight LLM QA protocol inspired by language-learning assessment, which better preserves salient content, aligns more closely with human rankings, and reveals differences between gloss-free and gloss-supervised systems that BLEU-4 obscures.
By Oline Ranum, Edward Fish, Simon Hadfield, Richard Bowden
ENEAS is a unified, text‑promptable method that simultaneously provides precise instance tracking and high‑quality segmentation, and enables open‑concept discovery of any instance named by a text query. It extends the SeC architecture with a text‑prompting adapter and temporal memory to maintain targets through disappearance and avoid drifting, while a semantic verification layer combines visual embedding matching with conditional VLM refinement to filter ontological errors. Designed for 3D reconstruction, ENEAS delivers robust semantic tracking and segmentation across videos, libraries, and unordered collections, distinguishing true instances from look‑alike doppelgangers.
By Javier del Pino (SperidLabs), Salvador Rodr\'iguez (SperidLabs), Alejandro Garabito (SperidLabs), Javier \'Alvarez (SperidLabs), Chema Garabito (SperidLabs)
The paper audits zero‑shot vision‑language models—CLIP, BioCLIP, BioCLIP2, and a multilingual Jina CLIP v2—on 10,321 images of seven Bangladeshi freshwater fish species. BioCLIP2 achieves the highest accuracy (72.36% on BFF‑15 with English common names and 68.91% on SylFishBD with scientific names), while generic CLIP lags far behind (25.15% and 14.40%). The study also shows that Bengali prompts perform poorly, that Jina CLIP partially improves Bengali discrimination, and that image context (blur, masking) affects performance differently across species.
By Nazim-E-Alam, Tarek Rahman, Md Kishor Morol
The paper introduces MINT‑Safe, a new open‑source dataset of 11,270 multi‑image dialogues and 500 refusal VQA pairs designed to expose safety risks in multi‑modal large language models during open‑ended conversations. It also proposes TAD‑Align, a turn‑aware dual‑objective reward framework that dynamically up‑weights dialogue turns with inconsistent safety behavior, improving safety metrics on Qwen2.5‑VL‑7B‑Instruct and LLaVA‑Next‑7B. The results show over 10% reduction in attack success rate and notable gains in harmlessness and helpfulness while maintaining overall model performance.
By Han Zhu, Jiale Chen, Chengkun Cai, Shengjie Sun, Haoran Li, Yujin Zhou, Chi-Min Chan, Pengcheng Wen, Lei Li, Yike Guo, Sirui Han
FailBench is a new benchmark for robot failure detection, containing 2,197 manipulation attempts from 14 public sources, with 75% of failures occurring naturally. The study evaluates 13 vision‑language model (VLM) detectors, finding the best model achieves only 0.77 mean balanced accuracy, and that fine‑tuned failure detectors often underperform general‑purpose VLMs. Performance varies with visual evidence, excelling when object motion is observable but dropping to near chance on contact‑intensive assembly tasks, and input‑level cropping of outcome‑relevant regions improves the top detector by 2.4 percentage points.
By Zaruhi Navasardyan, Tatul Danielyan, Hrant Davtyan
Puffin-World is a unified multimodal architecture that integrates physical understanding, spatial simulation, and 3D world generation without external offline modules. It jointly models physics, geometry, and appearance as native world states and uses a unified Omni-Camera representation to support diverse tasks and flexible motions. The framework also propagates physical dynamics across future frames, couples appearance and geometry in a single generative process, and scales to complex scenarios with the Puffin-16M dataset of 15 million vision‑language‑camera triplets and 1 million trajectories.
By Kang Liao, Yihang Luo, Xiao-Ming Wu, Linyi Jin, Size Wu, Chunyu Lin, Yao Zhao, Fei Wang, Wei Li, Chen Change Loy
The paper introduces GBA-GCs, a large-scale multimodal benchmark for identifying gated and open residential compounds in China’s Greater Bay Area, comprising 37,444 compounds with satellite imagery, metadata, and verified labels. It presents MCGC, a vision-centric multimodal framework that fuses imagery, text, and structured data to accurately classify gated communities, outperforming existing baselines. Using the model, the authors map gated communities across the metropolitan area and uncover equity-related patterns such as clustered gated zones, privatized green space, and diminished pedestrian connectivity.
By Minwei Zhao, Weiming Zhang, Jiawang Du, Qiming Liu, Weiming Zhuang, Pei Nie, Cai Wu
The paper introduces MedQA-MM, a benchmark that exposes shortcut reasoning in medical multimodal multiple-choice questions. By auditing prompts, images, and modalities, the authors show that models often rely on textual cues rather than visual evidence, with full-input accuracy at 62.63% but only 5.21% when restricted to text. The study highlights the need for route-level evidence to validate true medical image reasoning.
By Benlu Wang, Yifan Zhang, Jiaqing Yu, Chin Siang Ong, Juncheng Huang, Zhuohao Li, Zhenyu Zhang, Arman Cohan, Hong Yu, Zonghai Yao
The paper introduces TopKSigLIP, a vision‑language model tailored for mammography that tackles two key challenges: high‑resolution imaging and homogeneous radiology reports. It replaces standard CLIP training with a TopK‑Patch module that selects sparse high‑resolution patches likely to contain lesions, and a Sup‑sigmoid loss that uses soft labels from structured data instead of contrastive loss. TopKSigLIP outperforms existing open‑source mammography and general medical VLMs on zero‑shot tasks such as density assessment, BI‑RADS classification, finding subtyping, and cancer prediction, while also providing better lesion localization than Grad‑CAM.
By Young Seok Jeon, Beatrice Brown-Mulry, Rohan Satya Isaac, Anjana Dissanayaka, Theo Dapamede, Mohammadreza Chavoshi, Judy Gichoya, Hari Trivedi
AdaRoboVLG is a Vision‑Language‑Grasp framework that separates a generalizable base grasp policy from task‑specific understanding. The base policy generates and evaluates physically feasible grasp candidates using kinematic mapping and force‑closure stability, while foundation‑model modules supply composable spatial, cognitive, and temporal priors that adapt grasp synthesis to different robotic hands and environments without retraining. Experiments show strong cross‑hand generalization, effective handling of diverse grasping challenges, and functional grasping in cluttered, dynamic settings.
By Sixu Yan, Shikang Wang, Binhua Huang, Xuanlai Tang, Guohua Fan, Fan Huang, Haoxuan Li, Yongkang Li, Yuhan Li, Bencheng Liao, Zeyu Zhang, Wenyu Liu, Hangxin Liu, Xinggang Wang