7591 papers
HiDrop is a novel framework that significantly accelerates Multimodal Large Language Models (MLLMs) by aligning token pruning with hierarchical layer functions through Late Injection, Concave Pyramid Pruning, and Early Exit mechanisms, achieving a 90% reduction in visual tokens with a 1.72x training speedup while maintaining original performance.
This paper proposes a reliable indoor navigation system for humans that integrates Vuforia Area Target for environment modeling, AI NavMesh for pathfinding, and the A* algorithm to deliver faster, more accurate, and intuitive real-time guidance compared to traditional signage and GPS-based methods.
EgoGraph is a training-free, dynamic knowledge graph framework that overcomes the limitations of existing methods in ultra-long egocentric video understanding by constructing a unified schema and temporal relational modeling to capture long-term cross-entity dependencies, thereby achieving state-of-the-art performance on long-term video question answering benchmarks.
This paper introduces VGUBench to demonstrate that while Unified Multimodal Large Language Models exhibit strong textual reasoning and visual rendering capabilities individually, they fail to maintain semantic equivalence when required to generate visual answers, revealing a critical breakdown in cross-modal semantic alignment rather than a lack of generation fidelity.
StemVLA is an open-source Vision-Language-Action model that enhances robot manipulation performance on long-horizon tasks by explicitly integrating predicted future 3D spatial geometry and aggregated 4D historical spatiotemporal representations to improve spatial reasoning and decision-making in dynamic environments.
This paper proposes a novel difference-in-difference method that improves AI-generated image detection by utilizing second-order differences in reconstruction error to achieve superior generalization and accuracy compared to existing first-order approaches.
UTPTrack introduces a simple, unified token pruning framework that jointly compresses the search region and both dynamic and static templates via an attention-guided strategy, achieving state-of-the-art accuracy-efficiency trade-offs in visual tracking while preserving baseline performance across RGB and multimodal scenarios.
U-Mind is a pioneering unified framework that enables real-time, high-intelligence multimodal interaction by jointly modeling language, speech, motion, and video synthesis through a novel alignment and reasoning strategy to achieve coherent, synchronized, and expressive conversational agents.
This paper investigates the behavioral gaps between humans and Vision-Language Models in recognizing ambiguous Japanese characters by demonstrating that their decision boundaries differ in shape-only tasks, though contextual information can partially improve human alignment in VLMs.
This paper proposes CausalProto, an unsupervised causal prototypical network that leverages structural causal modeling and information bottleneck constraints to disentangle pathological features from environmental confounders, thereby achieving de-biased, interpretable, and high-accuracy dermoscopy diagnosis without compromising performance.
Neural Image-Space Tessellation (NIST) is a lightweight, screen-space post-processing technique that uses multi-scale neural operators to deform image contours and reassign appearance information, effectively simulating the visual fidelity of geometric tessellation on low-polygon meshes with constant computational cost independent of scene complexity.
The paper introduces Selfment, a fully self-supervised framework that achieves state-of-the-art object segmentation and zero-shot generalization to camouflaged objects by iteratively refining self-supervised patch features to generate high-quality pseudo-labels without any manual annotations.
This paper introduces OPTIAGENT, a physics-driven agentic framework that leverages Large Language Models enhanced with a specialized dataset, hybrid training objectives, and a physics-guided reward system to automate the design of functional lens systems, effectively bridging the gap between human expertise and automated optical engineering.
The paper introduces VideoPulse, a comprehensive dataset and end-to-end deep learning pipeline that enables accurate, contact-free estimation of neonatal heart rate and SpO2 from facial video, offering a low-cost, non-invasive alternative to traditional adhesive monitoring methods in intensive care settings.
This paper proposes a novel few-shot 3D vessel segmentation framework that adapts the pre-trained DINOv3 foundation model with specialized 3D components to achieve superior performance and robustness in data-scarce and out-of-distribution clinical scenarios, significantly outperforming state-of-the-art methods like nnU-Net with only five training samples.
This paper introduces FluoCLIP, a stain-aware focus quality assessment framework supported by the new FluoMix dataset, which addresses the limitations of existing stain-agnostic models by leveraging a vision-language approach to accurately evaluate focus quality across diverse fluorescent stains and tissue types.
The paper proposes EMO-R3, a reflective reinforcement learning framework that enhances the emotional reasoning capabilities of Multimodal Large Language Models by introducing Structured Emotional Thinking and a Reflective Emotional Reward to improve interpretability and alignment with human emotional cognition.
The paper proposes BiM-GeoAttn-Net, a lightweight framework that combines Bidirectional Depth Mamba for efficient cross-slice modeling and a Geometry-Aware Vessel Attention module to achieve robust, high-accuracy 3D segmentation of aortic dissection lumens in CTA scans.
The paper proposes Sea, an unsupervised cross-domain adaptation framework that employs a VLM-guided agent to actively navigate and select optimal viewpoints for frozen perception models, thereby significantly improving performance on tasks like visual grounding, segmentation, and 3D box estimation without requiring downstream labels or model retraining.
This paper proposes a bimanual manipulation framework that leverages a pre-trained 3D geometric foundation model to fuse RGB-based 3D latents, 2D semantics, and proprioception within a diffusion policy, enabling the joint prediction of actions and future 3D scene evolution to achieve state-of-the-art performance without relying on explicit point clouds.