The paper investigates how to close the quality gap in low‑resource text‑to‑speech for Khmer and Korean using the VoxCPM2 model. By training a single low‑rank adaptation (LoRA) adapter on a shared 25.5‑hour corpus, the authors improve Khmer’s mean opinion score from 3.85 to 4.23 with a rank‑64 adapter, while Korean shows no significant gain. The study highlights that adaptation benefits mainly when the base model is weak and that training loss does not always align with human ratings.
By Phannet Pov, Hyun Woo Park, Voneat Pen, Sovandara Chhoun, Wan-Sup Cho, Saksonita Khoeurn
MEVL-STP introduces a two‑stage pipeline for spotting arbitrarily shaped scene text. The detection stage fuses features from six frozen vision encoders via a hierarchical Feature Pyramid Network and a Progressive Scale Expansion network to produce precise polygon masks. The recognition stage then crops these masks and feeds them to a fine‑tuned Qwen3‑VL‑8B‑Instruct model, achieving state‑of‑the‑art detection and end‑to‑end performance on CTW1500, Total‑Text, and ICDAR 2015 without synthetic pretraining.
By Aman Anand, Partha Pratim Roy, Shivakumara Palaiahnakote
The paper introduces MK‑FSS, a few‑shot segmentation framework that leverages Multimodal Large Language Models (MLLMs) to extract spatial and semantic target knowledge from query images. Spatial knowledge is encoded into a memory representation and fused with support‑guided memory via a dual‑memory debate‑fusion module, while semantic knowledge is turned into a textual feature and combined with multi‑scale query features through a progressive cross‑modal prompt generator. Together, these components produce a robust target representation that improves segmentation performance over existing methods.
By Yijun Hu, Heng Fan, Libo Zhang
MoVISA introduces Multi-Token Reasoning for Video Object Segmentation, using multiple segmentation tokens (e.g., SEG0, SEG1) instead of a single token to better localize multiple objects over time. This approach enhances fine-grained alignment between language prompts and spatio-temporal mask predictions, leading to improved performance and interpretability. On benchmarks such as MeViS, DAVIS17, ReVOS, and Ref-Youtube-VOS, MoVISA achieves significant gains, notably a 13.2% J and F improvement on MeViS and an 8.4% J and F improvement on ReVOS.
By Ruining Zhao, Ho Kei Cheng, Alexander G Schwing
The paper introduces a tool‑augmented framework that enhances a small Vision‑Language Model (Qwen3.5‑4B) with geometric tools—3D object detection, metric depth estimation, and deterministic solvers for distance, size, and bearing—to improve metric spatial reasoning. By moving metric computation from the model’s weights into explicit solvers, the approach achieves significant gains on ReVSI‑Bench tasks, notably increasing absolute distance accuracy from 0.46 to 0.74 MRA and relative direction accuracy from 25.9% to 73.4%. The modular design allows swapping in different detectors, enabling a clear separation between perception and reasoning errors, and the model can autonomously sequence the tools to match a scripted pipeline on most tasks.
By Kai Glantz, Clemens Grange
IronViT proposes a new approach to building efficient generalist vision encoders by first consolidating the knowledge of multiple specialist teachers into a softmax attention bridge and then transferring this consolidated representation to a hybrid softmax‑linear attention architecture. This two‑stage distillation process, supported by a curated data pipeline, allows the model to capture semantic, spatial, language‑aligned, and action‑relevant cues while avoiding the high‑resolution cost of traditional softmax attention. Across tasks such as recognition, retrieval, dense prediction, multimodal understanding, and robotic learning, IronViT matches or exceeds the performance of leading specialist and generalist encoders, with the hybrid encoder offering increasing efficiency at higher resolutions.
By Jiaxi Huang, Yueqi Hu, Xin Zhu, Xiaopeng Zhang, Huiting Qiao, Yanglin Zhang, Zefeng Ji, Rongxue Li, Yifei Xu, Huiying Yu, Wei Liu, Jiayin Zheng, Yinggan Xu, Peipeng Chen, Yin Zhang, Jian Yao
The SEE Challenge 2026 invites participants to restore RGB images using synchronized event camera data and a target brightness statistic across a wide illumination range. Using the SEE-600K dataset of 610,126 image‑event pairs from 202 real‑world scenes, teams compete under an open‑system protocol, with PSNR as the primary ranking metric and SSIM as a secondary measure. Fifteen valid submissions were evaluated, revealing closely spaced top scores and consistent local errors under severe underexposure, while the report also examines exposure subsets, semantic test cases, shared failure patterns, and system design choices.
By Yunfan Lu, Mingchao Xu, Hanyu Zhou, Shaoyu Liu, Haoyue Liu, Peiqi Duan, Shihan Peng, Yinqiang Zheng, Boxin Shi, Gim Hee Lee, Hui Xiong, Davide Scaramuzza
The paper introduces Anomaly‑LR, a defect‑grounded latent reasoning framework for industrial anomaly detection that builds a global understanding of an image and then refines anomaly‑relevant representations directly in visual latent space. It also presents IAD‑LR‑22K, a new instruction dataset with 22,228 image‑question pairs and detailed annotations. Experiments demonstrate that Anomaly‑LR outperforms comparable‑scale methods on multiple IAD benchmarks without needing external references or tools.
By Jaron Yeh, Yen-Wei Chang, Jiang Liu, Shao-Yuan Lo
The paper introduces a vision‑language framework that extracts social indicators from street‑level imagery, converting panoramic views into sidewalk‑facing sideviews with timestamps. Using a VLM‑based activity detection system, it codes each pedestrian across ten observable dimensions, producing a Social Dwelling Index (SDI) that captures grouping, dwelling, activity diversity, and accessibility flags. Applied to over 100,000 sideviews in New York City, the study finds that pedestrian volume and SDI are only weakly correlated, indicating that high foot traffic does not necessarily equate to intense social activity.
By Liu Liu, Andres Sevtsuk
The paper introduces Self‑Adaptive VLA, a post‑training method that lets Vision‑Language‑Action policies self‑adapt to deployment‑time hardware shifts by using rollouts as context. It creates shift‑conditioned expert demonstrations, compresses visual, proprioceptive, and action data into a latent context token, and modulates the policy via adaptive layer normalization. Experiments on four precision‑critical manipulation tasks show the method recovers over 80 % of the base policy’s performance under actuation bias and encoder offsets, and improves robustness on new workstations.
By Hongxin Zhang, Chunru Lin, Tsun-Hsuan Wang, Zhenjia Xu, Chuang Gan
COMPASS is a completion-and-fusion framework designed for multimodal human activity recognition (HAR) and human pose estimation (HPE) when some modalities are missing at inference. It assigns each modality to a fixed slot, filling it with either an observed representation or a completion inferred from available inputs, and uses fusion‑matched supervision to train completions against real targets at the readout level. Experiments on XRF55 and MM‑Fi datasets show that COMPASS outperforms strong baselines and alternative matching strategies for both HAR and HPE.
By Hao Wang, Yanyu Qian, Pengcheng Weng, Zixuan Xia, William Dan, Yangxin Xu, Fei Wang
The paper introduces Hyperbolic Multimodal Continual Learning (HMCL), a method that preserves the Lorentz geometry of hyperbolic multimodal models during sequential updates. By restricting all modalities to a shared hyperbolic isometry, HMCL formulates a joint closest‑admissible (CA) correction—along with a minimal‑rotation (MR) variant—to adjust AdamW updates while maintaining task performance. Experiments on a 16‑task classification‑retrieval stream with three hyperbolic backbones show that HMCL-CA achieves the highest overall score, reduces geometric drift by up to 95.5 %, and improves semantic hierarchy preservation on ImageNet‑WordNet.
whyItMatters":"The study demonstrates that explicitly maintaining hyperbolic geometry during continual learning yields superior performance and reduced representation drift compared to existing baselines."
By Jiahong Liu, Ming Shen, Xiaohao Liu, Rex Ying, Menglin Yang, Tat-Seng Chua, Irwin King
The paper introduces TRACE, an amortized temporal gradient‑inversion attack that reconstructs private observation‑action trajectories from per‑step policy gradients in embodied reinforcement‑learning agents. TRACE exploits cross‑time correlation between gradients and exact action recovery from policy‑head gradients, achieving high reconstruction quality (18.8 dB PSNR) and near‑perfect action recovery with minimal computation. The study demonstrates TRACE’s effectiveness across various neural architectures and input modalities, and suggests that protecting temporal gradient streams may require sequence‑aware privacy mechanisms.
By Sudip Bhujel, Shanghao Shi, Ruiquan Huang, Ning Zhang, Yang Xiao
The paper introduces DEEPO, a Dual-Entropy Enhanced Policy Optimization method designed to mitigate hallucination in multimodal large language models (MLLMs). It addresses two weaknesses in reinforcement learning: (1) hard queries with high semantic entropy produce uniformly wrong samples, erasing advantage signals, and (2) confident-but-wrong tokens become invisible to gradients as the policy sharpens. DEEPO combines semantic‑entropy‑triggered expert prefixes to inject grounded continuations and Renyi preconditioning to counter logit saturation, yielding significant hallucination reduction while maintaining accuracy and training stability.
By Yingxuan Zhuang, Miao Pan, Wangjie Gan, Jingxiao Yang, Fan Wang, Weiming Liu, Cheng Tan, Xuhong Zhang, Jintao Chen
The Pistis Technical Report introduces the Pistis model family, comprising 27B- and 9B-parameter multimodal large language models built on Qwen3.6 and Qwen3.5. The models are developed through a scalable post‑training framework that begins with large‑scale multimodal supervised fine‑tuning and then applies Interleaved Distillation and Reinforcement Learning (IDRL) to integrate on‑policy distillation and reinforcement learning within a single training loop. Two specialized variants—Pistis‑Thinking for deep multimodal reasoning and Pistis‑Agentic for long‑horizon planning, iterative reasoning, and tool use—are produced at both scales, and a system‑level method called Pistis‑Auto‑Harnessing (PAH) further improves inference harness performance without updating model parameters.
By Heyun Chen, Xiaohan Lan, Jiaxi Li, Zhilin Lu, Qi She, Weiwen Xu, Fei Yu, Yujie Zhong, Jinghuan Chen, Zijian Feng, Siyu Jiao, Yiheng Lin, Xinhao Wang, Sihan Yang, Jieyu You, Changbin Zhang, Hengyu Zhang, Xudong Zhang, Yunqing Zhao, Shuai Zheng
SkinAgent AI is a multimodal, safety‑grounded framework designed for non‑diagnostic skincare support. It routes visual concerns (Acne, Pores, Wrinkles), estimates skin type from photographs, and provides count‑informed acne severity, all while grounding recommendations in a database and enforcing deterministic safety, privacy, and evidence checks. The system achieved high accuracy in routing and skin‑type estimation, and demonstrated no safety or privacy violations in controlled tests, though tool‑selection errors and incomplete product grounding remain.
By Muhammad Muhtasim Shahriar, Abdullah Mohammad Sayem, Tze Hui Liew, M. F. Mridha, Md. Mahiuddin
Jev-Mobile introduces a new approach for mobile GUI agents by separating high‑frequency lightweight execution from low‑frequency vision‑language model (VLM) planning. The VLM sets local goals, the accessibility tree provides a structured action space, and Jev, a fast typed decision model, repeatedly selects actions within this space, allowing multiple GUI actions per VLM decision. On the AndroidWorld task suite, Jev-Mobile achieves 79% task success, reduces mean end‑to‑end execution time by 32.7%, and cuts model API cost by 73.4% compared to a step‑wise VLM baseline.
By Linghua Zhang
Cross-Country Code-Mixing for Generative Recommendation (CMRec) is a framework that enhances generative recommendation across different countries by injecting cross-country supervision at the data level. It learns a shared semantic codebook from multi-modal content and behavioral co-occurrence, then synthesizes mixed-country sequences through token-level substitutions that respect both static and dynamic constraints. A context-aware loss reweights these mixed samples based on their plausibility, leading to improved recommendation quality in data-sparse countries while maintaining performance in data-rich markets, as demonstrated by significant gains in advertising revenue and orders in real-world e-commerce experiments.
By Yuan Gao, Hao Deng, Haibo Xing, Yi Xu, Lingyu Mu, Jinxin Hu, Yu Zhang, Xiaoyi Zeng
Med-AR introduces two autoregressive vision‑language models, Med‑AR‑8B and Med‑AR‑2B, pretrained on structured radiology reports, abnormality‑focused text, and region annotations to address long‑tailed chest X‑ray classification. The models outperform existing contrastive, self‑supervised, and supervised encoders—including Med‑CLIP, CheXFound, EVA‑Base, ARK, and BioViL‑T—across PadChest, MIMIC‑CXR, and CheXpert, achieving higher mean AUROC and AUPRC for head, medium, and tail findings and lower excess area under the risk‑coverage curve. Med‑AR also demonstrates improved selective‑prediction performance, with Med‑AR‑8B raising tail‑label mean AUPRC on MIMIC‑CXR from 0.1033 to 0.1441 and Med‑AR‑2B delivering the strongest discrimination on PadChest.
By Janhavi Prabhu, Sahil, Akshay V, Shivam Shukla, Manoj Tadepalli, Preetham Putha
Deep learning framework GLAM predicts glaucoma progression rates from longitudinal Humphrey 24‑2 visual field data and five clinical features, achieving a mean absolute error of 0.139 dB yr⁻¹ and an AUC of 0.990 for fast‑progressor detection. Using attention‑based fusion and aleatoric uncertainty, GLAM outperforms a ridge regression baseline by 73.5% in MD‑rate prediction. The model demonstrates that visual field data alone can match multimodal pipelines for progression prognostication.
By Taiabur Rahman, Siddiqur Rahman, Muhammad Moniruzzaman, Ummay Kawsar, Sayedatunnessa Ratna, Shadman Siddique, Rafsan Siddique, Tausif Ahmad, Tahsin Ahmad, Golam Rabbani