Agentic Video Intelligence: A Flexible Framework for Advanced Video Exploration and Understanding

The paper introduces SMART, a new framework for Video Moment Retrieval that enhances the localization of specific temporal segments in untrimmed videos using natural language queries. Traditional methods have limitations due to their reliance on coarse temporal understanding and single visual modalities. SMART addresses these issues by integrating audio cues and leveraging shot-level temporal structures, thus improving multimodal representations through Shot-aware Token Compression, which retains high-information tokens to enhance performance.

The paper introduces Risk Semantic Distillation (RSD), a novel framework aimed at enhancing end-to-end autonomous driving (AD) systems. While current AD systems perform well in complex scenarios, they struggle with generalization to unseen situations. RSD leverages Vision-Language Models (VLMs) to improve training efficiency and consistency in trajectory planning, addressing challenges posed by hybrid AD systems that utilize multiple planning approaches. This advancement is crucial for the future of autonomous driving technology.

MedGEN-Bench is a newly introduced multimodal benchmark aimed at enhancing medical AI research, particularly in the context of Vision-Language Models (VLMs). It addresses significant limitations in existing medical visual benchmarks, which often rely on ambiguous queries and oversimplified diagnostic reasoning. MedGEN-Bench includes 6,422 expert-validated image-text pairs across six imaging modalities and 16 clinical tasks, structured to improve the integration of AI-generated images into clinical workflows.

The article discusses the limitations of multimedia documents, which are often distributed in static raster formats, hindering their editability. To address this, a new framework called SliDer is introduced, utilizing Vision-Language Models (VLMs) to convert slide images into editable Scalable Vector Graphics (SVG) representations. This approach aims to preserve the semantic structure of documents, overcoming the shortcomings of traditional raster-vectorization methods that fail to maintain the distinction between image and text elements.

The article discusses advancements in fine-tuning Vision-Language Models (VLMs) to enhance spatial reasoning. Traditional methods often suffer from biases and errors due to imbalanced data collection and annotation from real-world scenes. To overcome these issues, the authors propose a redesigned fine-tuning process that includes controlled data generation and annotation, ensuring quality and balance. This approach involves comprehensive sampling of object attributes and aims to improve the transferability of VLMs to real-world applications.

DocLens is a newly proposed tool-augmented multi-agent framework designed to enhance the understanding of long visual documents. Traditional Vision-Language Models (VLMs) face challenges in evidence localization, often failing to retrieve relevant pages and missing fine-grained details, which can lead to model hallucination. DocLens addresses these issues by effectively zooming in on evidence and utilizing a sampling-adjudication mechanism to produce reliable answers. When paired with Gemini-2.5-Pro, it achieves state-of-the-art performance on benchmarks, outperforming even human experts.