Tag
This paper introduces M3R-Bench, a unified evidence-grounded benchmark for multimodal metaphor understanding with 1,000 image-text instances, and proposes M3R-Reasoner, an 8B-parameter model combining curriculum-based reasoning supervision and reinforcement learning that outperforms larger proprietary models.
This paper proposes a commutation theory for label-free reliability in vision-language figure reading, showing that consistency-based methods have a computable blind spot and introducing an Equivariance-Consistency Score enhanced by cyclic relabeling.
NVIDIA releases Nemotron Parse 2.0, a document image parsing model that converts scanned PDFs and images into structured text with layout, bounding boxes, and reading order, adding multilingual OCR improvements and chart-aware parsing.
Introduces CARGO-VL, a group-relative optimization framework for vision-language models that improves handling of conflicting image-text evidence and unsupported-answer avoidance via counterfactual consistency and risk-constrained control, along with the XMC conflict training resource.
ChronoVision is a multimodal framework that improves temporal reasoning in vision-language models by aligning visual logic with latent imagery, using a Reconstructive Visual Head, ROI Attention Locating module, and reinforcement learning. It introduces the Vbvr-VQA dataset and achieves SOTA accuracy on temporal tracking benchmarks.
This repository provides INT8 ConvRot-quantized ComfyUI safetensors of Qwen3-VL-32B, including a MiniMax-H3 conditioning encoder with layers 0-49 and an optional prompt-enhancement tail for layers 50-63, designed for use in ComfyUI on 32GB GPUs.
This paper introduces an information-asymmetric spot-the-difference task to measure epistemic vigilance in vision-language models, finding that models often overlook private evidence to agree with partners. Model steering to reduce sycophancy improves reliability in cooperative tasks.
This paper describes a two-stage vision-language adaptation system for Nepali meme classification, using Qwen3-VL-8B-Instruct with LoRA fine-tuning and contrastive learning. The system achieved 2nd place in hate speech detection and 4th in sentiment analysis at the CHiPSAL 2026 shared task.
Introduces CAPEval, a caption evaluation framework that decouples coverage and precision, showing that coverage better predicts vision-language understanding performance while precision better predicts text-to-image generation performance.
DeepVoyager-VL proposes a long-horizon multimodal deep-search framework that integrates visual evidence into intermediate reasoning, using a multimodal event graph for data synthesis and fine-tuning without reinforcement learning, achieving strong performance across ten benchmarks.
3DZip is a three-stage token compression framework for 3D vision-language models that uses voxelization, diversity-guided anchor selection, and spatial constraint merging to reduce tokens while preserving spatial reasoning. It retains 94.7% of original performance with only 128 tokens and achieves 1.92x faster inference on 3DQA benchmarks.
This paper presents ReBA (Relax Within, Balance Across), a geometry-guided load-balancing method for vision-language mixture-of-experts models. It addresses modality-specific routing imbalance by balancing image and text separately and grouping correlated visual tokens, improving robustness across resolution and composition shifts.
Introduces Poplar, a scalable Specify-Render-Inspect pipeline for synthesizing human-centric image datasets, and releases Poplar-9K, a curated dataset of 9,401 image-text pairs with auditable inspection records.
Mage-VL is introduced as a multimodal AI model that handles images, text, and video understanding in a single model, enabling richer interactive applications.
This paper presents VLD-RAG, an agentic multimodal retrieval-augmented generation framework for question answering over long, visually-rich documents. It uses a page-preserving index and a verifier-guided agent workflow to improve cross-page evidence retrieval and reasoning, outperforming prior vision-based baselines on benchmarks like LongDocURL and MMLongBench-Doc.
RRS-10K is a benchmark dataset for evaluating vision-language models on rare remote sensing image interpretation, containing over 10,000 military-related images and multiple task formats. Evaluation of 52 models reveals moderate zero-shot performance and weaknesses in visual grounding and complex reasoning.
ProcAgent is a fully on-device, agentic, vision-based procedural assistant that uses a propose-and-verify architecture for real-time adaptive guidance on an NVIDIA Jetson AGX Orin. It supports both reactive and proactive modes with human-in-the-loop confirmation, achieving responsive interaction and positive user study ratings.
This paper investigates why multimodal LLMs fail on vision-centric tasks when visual evidence conflicts with language priors, introducing the WhatIfVis benchmark and showing that models often encode visual evidence but cannot reliably control their reliance on it.
This paper investigates the scaling properties of native multimodal pre-training, deriving compute-optimal model sizes and token counts under a fixed budget, and revealing distinct scaling behaviors for language and multimodal objectives.
VCSD removes the need for external teachers, privileged answers, or visual evidence in on-policy self-distillation by using content-erased control images to produce contrastive signals, consistently outperforming existing methods on vision-language benchmarks.