The paper introduces POPEN, a new framework that uses preference‑based optimization and ensemble methods to reduce hallucinations and improve segmentation accuracy in large visual language models, achieving state‑of‑the‑art results on multiple benchmarks.
TwigVLM introduces a lightweight “twig” module that prunes visual tokens early and enables self‑speculative decoding, achieving up to 154% speedup on long‑text generation while preserving 96% of original LVLM accuracy, as demonstrated on LLaVA‑1.5‑7B and other benchmarks.
This article presents HiddenDetect, a training‑free method that leverages refusal‑semantic vectors and layer‑wise activation analysis to detect jailbreak attempts in multimodal large language models, revealing distinct safety signals across text and image modalities and demonstrating strong performance on several LVLM benchmarks.
VisionReasoner introduces a reinforcement‑learning‑driven unified framework that simultaneously handles detection, segmentation, and counting tasks within a single model, achieving 29.1% higher COCO detection AP, 22.1% better ReasonSeg segmentation, and 15.3% improvement on CountBench, while requiring only 7,000 training samples and offering efficient multi‑target matching via batch computation and the Hungarian algorithm.
