CoDeFont is a new one-shot font generation framework that learns to synthesize target glyphs by transferring style from a single reference glyph while preserving the target’s structure. It achieves this through multi-level style‑content factorization and progressive style‑content reconciliation, creating factor‑specific representations and using a contrastive objective to separate style and content embeddings. The method is evaluated on Bengali and Latin scripts, showing consistent improvements over existing state‑of‑the‑art approaches for both seen and unseen font styles.
By Rejoy Chakraborty, Prasun Roy, Saumik Bhattacharya, Umapada Pal
SATOR is a source‑free, backpropagation‑free framework for test‑time adaptation of 3D point cloud models. It eliminates the need for predefined purge‑size candidates by computing token abnormality scores from attention and geometric graphs, then using an Otsu‑based threshold to partition tokens into retained and pruned sets. A keep‑ratio‑calibrated routing step selects the pruned prediction only when its entropy, adjusted by the keep ratio, is lower than that of the original prediction, achieving state‑of‑the‑art accuracy on three 3D corruption benchmarks.
By Inseok Kong, Geunyoung Jung, Jiyoung Jung
RLDF (Region-Level Deepfake) is the first dataset designed to evaluate deepfake detection on partial facial manipulations. It contains over 2 million partial face‑swap images generated with eight face‑swapping methods across seven facial regions, including both discrete (eye, nose, mouth, eyebrow) and composite regions. The authors benchmark 12 state‑of‑the‑art detectors on five cross‑dataset protocols, revealing a 10.5 % drop in AUC and 10 % drop in accuracy for region‑level detection, and propose a mask‑guided cross‑attention model as a baseline.
By Shukesh Reddy, Abhijit Das
GRASE is a new agentic framework for converting real-world video into editable 3D scenes while preserving geometry and layout. It treats captured multi‑view video as a persistent reference, revising intermediate hypotheses whenever rendered evidence conflicts with the capture. By iteratively refining architectural shells, asset placement, and geometry with cross‑view feedback, GRASE produces faithful, consistent scenes that outperform existing state‑of‑the‑art methods.
By Zhaorui Lin, Haohan Liu, Zhen Chen, Xiangru Huang
MYOPIA is a benchmark that evaluates tree-based reasoning in vision‑language models (VLMs) using four human‑refined decision trees for GTSRB, MNIST, CIFAR‑10, and MSCOCO. The study shows that tree traversal consistently underperforms zero‑shot prompting across GPT‑4o, Qwen‑VL Max, and LLaMA‑3.2 11B Vision, with significant gaps between class‑attribute recall and image grounding. Diagnostics reveal that early incorrect decisions and answer extraction losses drive this performance drop, and node‑level analysis links split entropy to lower accuracy.
By Islam Mesabah, Sary Elmansoury, Gerrit Gro{\ss}mann, Peter Neigel, Sebastian J. Vollmer
The paper introduces HLR-AVSceneQA, a benchmark that tests audio‑visual large language models on a hear‑localize‑reason hierarchy, assessing recognition of sounds, localization in egocentric and allocentric views, and reasoning about their relation to the visual scene. It also presents HLR‑LLM, a model that adds a binaural spatial audio branch to a strong audio‑visual foundation and is trained with a curriculum including chain‑of‑thought supervision for multi‑step spatial reasoning. Experiments show that existing AV‑LLMs are limited in spatial grounding and relational reasoning, while HLR‑LLM markedly improves spatial audio‑visual scene understanding without hurting semantic perception.
By Kim Sung-Bin, Lee Jung-Mok, Jinwoo Jung, Oh Hyun-Bin, Hyeonggon Ryu, David Harwath, Tae-Hyun Oh
Anchor3DGS introduces a pose‑free, feed‑forward 3D Gaussian Splatting framework that replaces pixel‑aligned Gaussian prediction with a compact anchor‑based representation. By placing 3D anchors guided by information entropy, the method concentrates primitives in visually complex regions while ensuring broad spatial coverage, and each anchor aggregates multi‑view features to generate local Gaussians. A lightweight feature enhancement module refines the coarse RGB output, resulting in state‑of‑the‑art novel view synthesis quality with an order of magnitude fewer Gaussian primitives than prior approaches.
By Sheng Ye, Jiangke Lin, Fudong Wang, Yi Yuan, Zhen-Hui Dong, Yong-Jin Liu
SCMerge is a training‑free token merging technique for Vision Transformers that jointly considers feature similarity and spatial consistency when pairing tokens. The merged tokens are processed within the encoder and later reconstructed onto the original patch grid during decoding. Experiments on ADE20K, Pascal Context, and Cityscapes demonstrate that SCMerge improves the balance between segmentation accuracy and computational efficiency compared to existing methods.
By Junfeng Ren (Southern University of Science and Technology), Yi Gong (Southern University of Science and Technology), Yuan Zeng (Shenzhen Technology University)
The Digital Typhoon Dataset V3 extends the longest typhoon satellite image collection to nearly 50 years, now including multispectral satellite data for multi‑channel modeling. The study benchmarks machine learning models across the Northern and Southern Hemispheres, finding that multi‑channel inputs do not consistently outperform single‑channel ones for pressure regression, with ResNet‑18, full‑resolution images, and Channel 1 delivering the best single‑channel results. Systematic bias analysis reveals that intensity regression underestimates strong typhoons, and imbalance‑aware weighting strategies fail to fully correct this bias.
By Asanobu Kitamoto, Victor Le Louarn, Noelia Falczuk, Ngoc Anh Tong, Anja Veleva
The study benchmarks foundation models for chest X‑ray classification across a domain shift from adult to pediatric patients. Six foundation models and two CNN baselines were evaluated on a harmonized five‑label task using adult data for adaptation and tested on both adult and pediatric sets. Results show all models experience a substantial performance drop—over 10 percentage points—when applied to pediatric data, and deeper adaptation does not markedly improve cross‑population generalization.
By Sara In\'acio, Hugo Proen\c{c}a, Jo\~ao C. Neves
The paper introduces ACT3D, a vision‑language model that learns to think and explore in 3D environments using only single‑image inputs and action‑based interactions. By fine‑tuning with trajectory supervision and then applying reinforcement learning driven by final‑answer correctness, ACT3D develops an exploration policy that adapts to the question and improves accuracy across six benchmarks, outperforming both proprietary and open‑source baselines while using fewer input frames. The study shows that outcome‑based RL can teach VLMs to decide when and how to acquire visual evidence, leading to better reasoning and spatial understanding.
By Yijiang Li, Bingyang Wang, Nuno Vasconcelos
The paper "Seeing is Believing: Benchmarking and Mitigating the Interference of Language Priors on Visual Perception in MLLMs" identifies that language priors can override visual evidence in multimodal large language models, leading to inflated accuracy metrics. It introduces ViSiP-Bench, a benchmark using rare-but-real scenarios across six prior categories, a dynamic prior induction protocol, and three metrics that more accurately gauge prior resistance. Evaluation of 24 models shows pervasive prior adherence, little improvement with scale, and a tendency for larger models to either strongly override or firmly adhere to priors. The authors propose MAPS, a fine‑tuning approach that softens excessive prior reliance by targeting each model’s strongest priors with factually grounded knowledge.
whyItMatters":"The study highlights that current accuracy‑based evaluations overestimate models’ visual perception due to language priors, and offers a more nuanced benchmark and mitigation strategy to improve multimodal model reliability."
By Li Ma, Yan Zhuang, Jiawen Deng, Fuji Ren
Flow-CLIP introduces a method to improve compositional image-text matching by refining global image-text alignment with image-conditioned changes in token–patch evidence. It combines Global Compatibility Calibration (GCC) for sentence-level similarity, Image-Conditioned Evidence Refinement (ICER) to incorporate visual patch context, and Cross-Image Margin Ranking (CIMR) to enhance ranking margins when images support correct descriptions. Experiments demonstrate that Flow-CLIP outperforms pretrained CLIP and other global alignment approaches on binding accuracy and reduces reliance on language priors.
By Minghong Wei, Enguang Zhu, Pu Cao, Lu Yang, Qing Song
Visual Sanad is a new benchmark that tests vision‑language models (VLMs) for false acceptance of Qur’anic verse attributions across 2,400 images derived from 100 English Qur’an verses. The dataset includes four factual states per verse and six visual conditions, from plain text to decorative layouts, and all pairs were source‑checked and human‑audited. Experiments on 24 model‑prompt runs show a sharp split: proprietary API‑served VLMs vary from near‑perfect verification to moderate false acceptance, while recent open‑weight VLMs often default to VALID and show weak discrimination, though de‑biased prompting can reduce false acceptance by shifting thresholds toward INVALID.
"whyItMatters":"The study reveals that fluent VLMs may treat displayed religious citations as self‑authenticating unless they possess strong citation‑verification ability, exposing a culturally grounded multimodal factuality failure."
By Niaz Mohaiman Abtahi, Reyana Shahid, Evan Maksud
The paper audits 12 vision‑language models on a 1,195‑image Arabic calligraphy benchmark, testing numeric, visual‑cue, and Diwani‑first prompts. Accuracy peaks at 60.9% for closed/API models and 47.2% for open‑weight models, but many models collapse to a single default label (often Thuluth) far above the majority‑class baseline. Prompting improves performance unevenly, with visual cues boosting some models by over 10 points while others suffer, and foregrounding Diwani often shifts but does not eliminate the default bias.
By Niaz Mohaiman Abtahi, Evan Maksud, Reyana Shahid
FeaTok introduces feature-loss tokenization for image generation, training discrete tokens by reconstructing pretrained features instead of pixels. This approach improves generative performance across multiple self-supervised models (SigLIP, SigLIP2, DINOv2, DINOv3). In class-conditional generation, FeaTok reduces gFDr6 by up to 29% at 256×256 resolution and outperforms the strongest unguided continuous baseline with 38× less sampling compute, while also achieving higher human-preference reward-model scores and surpassing text-conditional baselines after only 500 fine-tuning updates.
By Zhang Xiaofeng, Samuel Lavoie, Nicolas Beltran-Velez, Jianhao Yuan, Felix Friedrich, Jakob Verbeek, Aaron Courville, Michal Drozdzal, Adriana Romero-Soriano
The paper introduces SCCL-DT, a method that improves zero‑shot chest X‑ray diagnosis by addressing semantic conflicts in contrastive vision‑language pre‑training. It uses adaptive thresholds derived from exponential‑moving‑average statistics to hard‑mask conflicting pairs, and combines two teachers—an inclusive and a consensus rule—through distillation into a single student model. On five public chest X‑ray benchmarks, SCCL‑DT achieves a mean zero‑shot AUC of 0.850, outperforming the best baseline at 0.837.
By Ting Liu, Haoran Lai, HaobinChen, Ming Liu, S Kevin Zhou
OpenGCUA is an open‑stack project that aims to make generalist computer‑use agents (GCUs) fully accessible to the research community. The first release focuses on code execution, allowing agents to use intermediate results and programming skills directly in computer interaction, and introduces harness‑to‑model decision transfer to shift observation and task‑state decisions from fixed rules to the model. On the OSWorld‑v2 benchmark, the full configuration boosts task scores for all eight evaluated models—including GPT‑6 Astra and Claude Opus 5—by an average of +19.6 partial‑score points and cuts steps per task by 53 % compared to batched GUI‑only execution, while also providing a real‑device harness and cross‑device SDK.
whyItMatters":"By releasing the full stack, OpenGCUA enables systematic study and broader experimentation with generalist computer‑use agents, addressing the current limitation of closed implementations and demonstrating significant performance gains across diverse models."
By Hanzhang Zhou, Long Li, Quyu Kong, Lei Wang, Panrong Tong, Long Chen, Tianyu Xia, Gongjie Zhang, Xu Zhang, Chen Liu, Jianan Zhang, Chenglin Cai, Zhengwen Zeng, Steven Hoi, Yue Wang
SCOPE introduces a reinforcement learning approach for text‑to‑image prompt engineering that respects semantic spans such as objects, counts, attributes, and relations. By pooling policy log‑probability shifts within each semantic span, the method assigns a shared coefficient to token losses while using a single advantage per completion, eliminating the need for separate span rewards or extra image renders. Experiments on Qwen‑Image, SD3, and FLUX.1‑dev show average relative gains of 9.48% on GeneVal and 7.62% on T2I‑CompBench compared to vanilla GRPO, with ablation studies confirming the benefit of semantic boundaries and pooling.
By Yufei Jiao, Jinfa Huang, Zeyue Xue, Tianyu Xie, Yuexiao Ma, Rongrong Ji, Xiawu Zheng, Jiebo Luo
RUSplatting introduces a robust 3D Gaussian splatting framework for reconstructing high‑fidelity underwater scenes from sparse views. The method decouples RGB learning guided by underwater attenuation physics, employs a novel adaptive weighting scheme for frame interpolation, and adds a loss function that reduces noise while preserving edges. Experiments on the newly released Submerged3D dataset show PSNR gains up to 1.90 dB over state‑of‑the‑art methods, improving both visual quality and geometric accuracy for deep‑sea rendering.
By Zhuodong Jiang, Haoran Wang, Guoxi Huang, Brett Seymour, Nantheera Anantrasirichai