7541 papers
This paper presents AgilePruner, an adaptive visual token pruning framework for Large Vision-Language Models that leverages empirical insights into the complementary strengths of attention-based and diversity-based methods to reduce computational overhead while mitigating hallucinations across varying image complexities.
The MAMA-MIA Challenge establishes a large-scale, multi-institutional benchmark using US training and European test data to evaluate and improve the generalizability and fairness of AI models for breast MRI tumor segmentation and treatment response prediction across diverse demographic subgroups.
This paper proposes a method to improve fast diffusion-based computed tomography reconstructions from sparse data by incorporating a complementary imaging modality (such as X-ray CT) as a guidance signal without requiring retraining of the diffusion prior.
This paper presents a unified framework that synthesizes modular factor graph modeling with certifiable convex relaxation techniques by demonstrating that key mathematical transformations preserve factor graph structure, thereby enabling the use of existing, high-performance robotics libraries to implement globally optimal estimation without requiring specialized solver expertise.
FoSS is a dual-branch framework that integrates frequency-domain Fourier analysis with linear-time selective state-space models to achieve state-of-the-art trajectory prediction accuracy on Argoverse benchmarks while significantly reducing computational complexity and model parameters.
This paper proposes an uncertainty-aware multi-task framework for breast ultrasound analysis that leverages multi-level bidirectional decoder interactions and adaptive feature weighting to overcome task interference and improve simultaneous lesion segmentation and tissue classification performance.
This paper presents a controlled study demonstrating that reinforcement learning primarily sharpens output distributions and improves sampling efficiency in medical Vision-Language Models only after supervised fine-tuning has established non-trivial reasoning support, leading to a boundary-aware training recipe that achieves strong performance across medical benchmarks.
This paper presents AG-VAS, a novel zero-shot visual anomaly segmentation framework that leverages Large Multimodal Models enhanced with learnable semantic anchor tokens, a Semantic-Pixel Alignment Module, and a specialized instruction dataset to overcome limitations in abstract concept representation and achieve state-of-the-art localization performance across industrial and medical benchmarks.
This paper presents a comparative evaluation of supervised learning and open-vocabulary vision models for post-disaster scene understanding across multiple datasets, concluding that while foundation models offer flexibility, supervised training remains the most reliable approach for accurately detecting small objects and delineating boundaries in cluttered disaster scenes when annotations are available.
The paper proposes NVB-Face, a novel one-stage framework that directly generates consistent, high-quality novel-view face images from a single degraded (blind) input by extracting features and transforming them into 3D-aware latent representations via a diffusion model, thereby outperforming traditional two-stage restoration-and-synthesis pipelines.
The paper proposes Perspective-Equivariant Fine-tuning for Demosaicing (PEFD), a novel framework that enables high-fidelity, ground-truth-free multispectral image reconstruction by leveraging projective geometry and adapting pretrained foundation models to outperform existing methods on real-world datasets.
MixerCSeg is an efficient, state-of-the-art architecture for crack segmentation that integrates CNN-like local texture analysis, Transformer-style global dependency modeling, and Mamba-inspired sequential context processing within a unified encoder, enhanced by specialized edge-aware and multi-scale fusion modules to achieve high performance with minimal computational cost.
The paper proposes TIMI, a training-free framework that achieves high spatial fidelity in image-to-3D multi-instance generation by leveraging pre-trained model priors through an Instance-aware Separation Guidance module for disentanglement and a Spatial-stabilized Geometry-adaptive Update module for geometric preservation, outperforming existing methods without additional training overhead.
This paper proposes AOT, a training-free method that optimizes token reduction in Video Large Language Models by establishing local and global token anchors and aggregating informative contexts via optimal transport to efficiently eliminate redundancy while preserving spatiotemporal fidelity.
UniTalking is a unified, end-to-end diffusion framework that leverages Multi-Modal Transformer Blocks and pre-trained video priors to generate high-fidelity, lip-synchronized talking portraits with personalized voice cloning, achieving superior performance over existing open-source methods.
The paper introduces SeaVIS, the first online framework for audio-visual instance segmentation that utilizes Causal Cross Attention Fusion and Audio-Guided Contrastive Learning to effectively track and segment sounding objects in continuous video streams while suppressing silent instances.
This paper introduces LinkVLA, a novel Vision-Language-Action model for autonomous driving that unifies language and action tokens in a shared codebook, incorporates an auxiliary action understanding objective for bidirectional semantic alignment, and employs a coarse-to-fine generation strategy to significantly improve instruction following, driving performance, and inference efficiency.
This paper demonstrates that the effectiveness of global token mixing in MRI restoration is highly task-dependent, showing that while local gated CNNs suffice for reconstruction and super-resolution tasks constrained by physics or preserved low-frequency data, global models are superior for denoising tasks involving spatially heteroscedastic noise.
This paper introduces the Deepfake Forensics Adapter (DFA), a novel dual-stream framework that integrates a frozen CLIP model with global and local forensic analysis streams to achieve state-of-the-art generalization in deepfake detection, particularly demonstrating significant performance improvements on the challenging DFDC dataset.
The paper introduces VidDoS, the first universal Energy-Latency Attack framework for Video-LLMs that employs masked teacher forcing and refusal penalties to generate instance-agnostic triggers, causing extreme token expansion and inference latency that lead to critical safety violations in real-time applications.