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.
The paper introduces LatentStream, a progressive latent working memory framework for streaming video understanding. It replaces the traditional store‑and‑retrieve approach with a retrieve‑and‑internalize strategy, organizing visual history into short, mid, and long‑term levels and progressively expanding memory receptive fields to internalize evidence into a compact latent memory. The method also employs confidence‑guided optimization to refine memory tokens, achieving state‑of‑the‑art results on online and offline video benchmarks.
SVG-Score introduces a human‑aligned evaluation framework for text‑to‑SVG generation, addressing the inadequacies of existing image‑based metrics like CLIPScore that poorly capture SVG‑specific errors such as color, count, and spatial inaccuracies. The authors first demonstrate that CLIP‑based scores are largely insensitive to these errors and that off‑the‑shelf Vision‑Language Models respond unevenly across error types and styles. They then create a human‑annotated Semantic Alignment dataset and develop two evaluators: a CLIP scorer adapted to vector graphics and a VLM judge trained with supervised fine‑tuning and reinforcement learning, enabling both fast large‑scale and expressive, interpretable assessment of SVG generators.
The paper introduces a landmark‑based approach to classify injury‑associated athlete sessions using minute‑resolution multimodal football monitoring data. Instead of labeling every minute, it creates a single representation per athlete‑session at fixed time points (landmarks) based on information up to that moment, thereby aligning the session‑level injury label with the data. Experiments on 3,743 elite women’s football sessions show that cumulative and dynamic logistic regression models achieve ROC‑AUCs between 0.367 and 0.607, though results carry wide uncertainty.
The paper introduces a reverse sign language dictionary that recognizes signs from continuous signing without relying on gloss annotations. It does this by captioning sign-level video clips into free-form procedural descriptions using an open-weight vision‑language model, then retrieving the closest description from a multilingual sentence encoder’s vocabulary. Experiments on a Japanese Sign Language 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%, matching or surpassing traditional closed‑set classifiers where applicable.
KnowVis is a framework that converts linear video lectures into knowledge‑centric visual summaries. It first builds a detailed concept map from multimodal video content to identify key and challenging concepts, then organizes these into structured knowledge units before synthesizing engaging visual narratives. The authors also provide a dataset of 125 educational videos with 1,079 visual summaries and show through automated metrics and a human study that KnowVis outperforms existing methods in accuracy, clarity, and learning outcomes.
The paper introduces FactoSR, a factorized reinforcement learning framework designed to improve spatial reasoning in Vision‑Language Models (VLMs). By decomposing the problem into planar correspondence (XY), depth consistency (Z), and temporal reversibility (T), FactoSR addresses the dimensional mismatch between 2D visual inputs and 3D physical reasoning. Experiments on multi‑view and video benchmarks show significant performance gains, with a 5.9% improvement on VSI‑Bench and 4.5% on All‑Angles‑Bench.
The paper investigates how audio‑video diffusion models use cross‑modal attention, focusing on the "attention triangle" that connects text, audio, and video streams. It finds that the audio‑video edge is bidirectional and heavily influenced by model biases, leading to semantic leakage when prompts conflict with learned priors. By extracting attention signals, the authors develop diagnostic tools and inference‑time interventions that improve cross‑modal alignment without sacrificing generation quality.
The paper introduces WIDE, a method for cross‑modal generative retrieval that tackles information asymmetry between text queries and visual candidates. WIDE uses Adaptive Entropy Thresholding to set uncertainty limits, Asymmetry‑aware Wildcard Decoding to emit wildcards where the model lacks fine‑grained detail, and Blind‑Spot Re‑ranking to score an expanded candidate set with both discrete confidence and continuous similarity. Experiments on the M‑BEIR benchmark show that WIDE reduces forced hallucination and outperforms existing generative retrieval approaches while keeping index structures compact.
Air-Ground Collaborative Vision-and-Language Navigation (AGC-VLN) pairs an unmanned aerial vehicle (UAV) with a global bird’s‑eye view and an unmanned ground vehicle (UGV) with a local first‑person view, creating a shared bird’s‑eye map that enables collaboration. The training‑free baseline decomposes navigation into VLM‑based semantic reasoning and deterministic geometric execution, allowing the UAV to render the UGV’s pose and target markers while the UGV plans a road‑following path using the shared map. In CARLA‑Air’s Town10HD scene, AGC‑VLN achieves a 77.0% joint success rate, a 27.0% improvement over the weaker individual agent and surpasses the strongest single‑agent baseline by 24.0 points.
The paper investigates how frozen vision‑language models (VLMs) respond to edits made directly to their internal object‑token representations, bypassing the image input. It introduces an answer‑key‑free protocol that evaluates edits by logical consistency and self‑audit, revealing that 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 free‑text VQA performance and even outperform oracle methods on remote‑sensing benchmarks, with findings consistent across multiple datasets and model backbones.
OmegaUse‑SOP is a human‑in‑the‑loop system that transforms expert demonstrations of professional computer use into reusable skills for graphical user interface (GUI) agents. It refines demonstrations through four modules—Observe, Reason, Configure, and Execute—to record multimodal traces, abstract them into semantic steps, incorporate domain rules, and execute the resulting skills with verification. In a collaboration with a power‑sector client, OmegaUse‑SOP was tested on photovoltaic simulation workflows in PVsyst 7.2, showing improved reliability for GUI agents on professional SOP tasks.
By Yixiong Xiao, Lang An, Hucheng Yang, Pinxue Ma, Yongquan Chen, Jingjia Cao, Yusai Zhao, Ting Wang, Ting Liu, Siqi Bao, Jingbo Zhou, Hua Wu
The study presents a multimodal BrainAGE framework that fuses structural T1-weighted MRI with DeepCBV maps—vascular information synthesized from non‑contrast MRI—to estimate brain age more accurately. Using two separate 3D convolutional neural networks, the combined model achieved a mean absolute error of 3.95 years on cognitively normal controls, outperforming single‑modality models. Saliency analyses showed MRI highlighted white matter and cortical atrophy, while DeepCBV emphasized vascular‑rich regions, and the integrated approach revealed stronger differentiation between stable and progressive MCI, indicating sensitivity to early vascular changes.
By Jordan Jomsky, Zongyu Li, Kay C. Igwe, Yiren Zhang, Max Lashley, Tal Nuriel, Andrew Laine, Scott A. Small, Jia Guo, for the Frontotemporal Lobar Degeneration Neuroimaging Initiative, for the Alzheimer's Disease Neuroimaging Initiative
MUDIDI is a two-stage framework designed to digitize multilingual dictionaries that are currently only available as scanned images. The first stage assesses character recognition and markup preservation, while the second stage segments dictionary entries and maps them into the SIL Multi-Dictionary Formatter schema. The authors also release a dataset of 30 annotated dictionaries and benchmark OCR, LLM, and VLM systems, finding that LLMs generally outperform others and that providing additional context improves digitization quality.
By David Setiawan, Temuulen Khishigsuren, Milind Agarwal, Pagnarith Pit, Aso Mahmudi, Ekaterina Vylomova
The paper introduces a new task called video object segmentation‑aware audio generation, which conditions sound synthesis on object‑level segmentation maps. It presents SAGANet, a multimodal generative model that uses visual segmentation masks, video, and textual cues to produce controllable audio for musical instruments, offering fine‑grained, visually localized control. The authors also release the Segmented Music Solos dataset of instrument performance videos with segmentation information to support this task and demonstrate that SAGANet outperforms current state‑of‑the‑art methods in controllable, high‑fidelity Foley synthesis.
By Ilpo Viertola, Vladimir Iashin, Esa Rahtu
FairLens is a benchmark and evaluation framework that measures fairness and validity of vision‑language models (VLMs) in high‑stakes domains such as hiring, legal, and healthcare. It uses over 100,000 face‑image and question pairs covering gender, race, and age, and assesses responses through demographic parity, soundness, demographic association, and bias in free‑text generation. The study finds that VLMs often make unwarranted inferences from faces rather than abstaining, especially in legal and healthcare contexts, and that small parity gaps can still hide unsafe treatment across groups.
By Vahid Reza Khazaie, Ahmed Y. Radwan, Shaina Raza
CACTUS is a new backdoor attack for decentralized federated learning that uses mask‑guided, modality‑specific operators to convert label‑consistent semantic pairs into target‑directed representation shifts. By isolating trigger effects and applying them counterfactually to clean embeddings before peer aggregation, CACTUS can propagate backdoors across repeated aggregation rounds. Experiments on speech, text, tabular, and image tasks show that with 30% malicious nodes, CACTUS achieves a mean attack success rate of 51.2% on Speech Commands and the highest mean ASR among evaluated attacks on three of four modalities, with success varying by network topology and malicious‑node ratio.
By Chao Feng, Burkhard Stiller
The paper introduces a black-box membership inference attack framework tailored for fine-tuned text-to-speech models, addressing challenges in query generation and representation engineering. It evaluates five query types, finding recitation queries most effective, and uses multi-level speech embeddings with temporal alignment for fine-grained comparison. Experiments on CosyVoice2, F5-TTS, and XTTS-v2 trained on VCTK and British Dialect datasets show high privacy leakage, with speaker-level AUC above 0.80 and record-level AUC between 0.80 and 0.90.
By Kunlin Cai, Kaiyuan Zhang, Zihang Xiang, Jinghuai Zhang, Abeer Alwan, Fnu Suya, Yuan Tian
The paper introduces DynaTokens, a transformer-based token generator that produces fine‑tuning tokens on demand for continual VideoQA with multimodal large language models. It uses shared generation weights and meta‑learning‑inspired regularisers to reduce task interference and forgetting, connecting the objective to sharpness‑aware optimisation for flatter cross‑task minima. Experiments on standard continual VideoQA benchmarks show that DynaTokens achieves higher average accuracy, lower forgetting, better zero‑shot generalisation, and robust cross‑modal transfer in a new ImageQA→VideoQA protocol.
By Toan Nguyen, Yang Liu, Celso De Melo, Flora D. Salim