§ Multimodal

Black Forest Labs releases FLUX 3 multimodal model with reported improvements over Gemini 2.0, Grok Imagine, and includes video-action robotics variant.

VLM-IE3D framework enhances vision-language models with implicit and explicit 3D geometry tokens from RGB video for improved spatial reasoning.

GraphVid enables precise multi-object video generation control via graph-structured representations instead of trajectory or text constraints.

MIRROR framework exploits complementary reasoning paths across text, diagram, and combined modalities to improve vision-language model reasoning on geometry problems.

X³-OPD cross-modal distillation framework transfers reasoning from text LLM teacher to audio-language student via on-policy alignment and acoustic perception.

ElasticTTT framework prevents prior collapse in test-time tuning of diffusion models for video editing by preserving distribution-mapping during optimization.

Speech-based multimodal LLMs detect cognitive impairment across diverse speakers and devices by leveraging linguistic and acoustic biomarkers with improved generalization.

Thinkink: 2D spatial interface integrating handwritten/sketch prompts with LLM responses via semantic tree interpretation.

Multi-axis evaluation framework for structured audio captions on AudioCards dataset validates five orthogonal dimensions beyond flat text metrics.

MSBraM: self-supervised foundation model for EEG capturing multi-scale temporal brain dynamics across downstream tasks.

ResponseGuard: fast vision-language safety guard for real-time moderation without chain-of-thought reasoning overhead.

Multimodal foundation model for EEG combines Mamba raw-signal encoder and ViT for time-frequency data to improve generalization.

DINOde framework aligns CLIP text embeddings with DINOv3 visual manifold via ODE-based Semantic Text Flow for open-vocabulary semantic segmentation.

M³-Gen uses GANs to generate gene expression profiles from histopathology images and clinical data, addressing biomedical data scarcity.

PC-Edit applies prompt-contrastive learning for precise region discovery in image editing, handling category/shape replacement with source removal.

OpenAI launches Health feature in ChatGPT enabling U.S. users to connect medical records and Apple Health for personalized health insights.

Vision-language models show consistent performance gaps based on image-vs-question modality order; test-time training method closes gap across benchmarks.

Condition Dropout (ConD) mitigates RGB-D semantic segmentation degradation when one sensor modality fails via continued training with dropout simulation.

Multi-modal transformer jointly processes time-series, wavelet, and spectral representations for nanopore blockade signal classification in single-molecule sensing.

18 medical image encoders on 650k radiographs show self-supervision drives representational convergence more than clinical labels.

Evaluates LLM-based machine translation on culturally loaded expressions using Dream of the Red Chamber Chinese-Japanese dataset; identifies systematic gaps.

Audio-Zero: label-free self-play framework for large audio language models improving fine-grained event and duration reasoning.

Appearance Pointers enable spatial region control in Diffusion Transformers via compact tokens, improving controllable image generation.

PathAgentBench: benchmark for vision-language agents on gigapixel whole-slide pathology images evaluating multi-scale evidence-seeking.

MeetingToM benchmark evaluates multimodal LLMs on theory-of-mind reasoning in multi-party meeting scenarios.

Vision-language models as unified backbone for attributed graphs with heterogeneous modalities (text, visual, mixed).

Dataset of human image similarity judgments across semantic aspects; benchmarks vision-language models on context-dependent perceptual metrics.

Domain-generalized pixel-level tampering detection robust across VLM-generated manipulations from ChatGPT, Gemini, Qwen-Image.

GigaPath-Flash and GigaTIME-Flash: computationally efficient pathology foundation models for whole-slide image and tumor analysis.

Benchmark evaluating LLMs' ability to reason about 3D spatial constraints in structure-based drug design vs. diffusion models.

Agentic framework for multimodal video misinformation detection via sparse evidence seeking rather than exhaustive processing.

Future-state-conditioned vision-language navigation trains policies to predict future visual outcomes beyond next-action supervision.

ActiveVision benchmark measures whether MLLMs perform active observation via 17 tasks requiring repeated visual perception.

ToolSciVer framework uses visual tools and RL to improve multimodal scientific claim verification from figures, tables, and charts.

HCIG framework uses hierarchical cross-modal graph networks to detect sarcasm and cyberbullying via text-visual incongruity.

DELUGE: multimodal deep learning system for continental-scale daily pluvial flood damage prediction at 1km resolution using foundation model embeddings.

SciForge: AI-native multimodal workbench for scientific discovery with agent-accessible services for code, datasets, workflow execution, and paper management.

Vision-language model achieves SOTA remote sensing benchmarks via simple scaling recipe without task-specific architectural changes.

PEACE framework transfers adult ECG models to pediatric populations via cross-modal alignment; addresses label scarcity.

SceneBind combines semantic and 3D spatial understanding across vision, audio, language with object-centric slots and uncertainty modeling.

MediaEval Medico 2025 analysis: parameter-efficient VQA achieves leaderboard gains but structured reasoning better ensures clinical interpretability.

TikStance: 13,876-comment multimodal dataset for stance detection across Trump, Biden, Harris in 2024 U.S. election TikTok discourse.

Symbal detects systematic misalignments in MLLM-generated image captions by identifying recurring errors tied to specific visual features.

VEXMLM extends XLM-R with language-specific tokenizers for Ge'ez-script languages (Amharic, Tigrinya) to reduce OOV rates in low-resource African NLP.

MM-IssueLoc benchmark isolates visual evidence impact in multimodal software repository issue localization across 23 languages and 652 instances.

Benchmark evaluates six MLLMs on scientific visualization literacy using 49 standardized assessment items across 8 visualization techniques.

Google Vids adds Gemini Omni support and personal avatar features for video generation and editing.

Thinking Machines Lab releases Inkling, a 975B-parameter open-weights MoE multimodal model trained on 45T tokens.

AlphaWiSE: post-hoc weight interpolation maintains cross-modal alignment in CLIP during continual multimodal learning via per-tensor scalar coefficients.

Bottom-up pipeline (YOLO + LayoutReader) and Tiramisu model for hierarchical structure understanding in dense newspaper images.