3882 papers
This paper proposes a reparameterized Tensor Ring functional decomposition that leverages Implicit Neural Representations and a structured basis combination to overcome the high-frequency modeling limitations of traditional methods, achieving superior performance in multi-dimensional data recovery tasks such as image inpainting and point cloud reconstruction.
FastLightGen is a novel algorithm that simultaneously compresses model parameters and reduces inference steps through an optimized teacher-student distillation framework, achieving state-of-the-art efficiency and visual quality in video generation with significantly fewer resources.
This paper introduces VSearcher, a reinforcement learning-based multimodal search agent that transforms static models into capable long-horizon web browsers through an iterative data synthesis pipeline and an SFT-then-RL training strategy, achieving superior performance on the proposed MM-SearchExam benchmark.
This paper introduces Think-as-You-See (TaYS), a unified framework that enables concurrent, streaming Chain-of-Thought reasoning for Large Vision-Language Models by decoupling visual encoding from textual reasoning, thereby outperforming traditional batch and interleaved approaches in both accuracy and latency for real-time video understanding.
CoEditor++ is a training-free, cognitively structured framework that leverages a two-stage "what-to-edit" and "how-to-edit" reasoning process with self-reflection to achieve state-of-the-art, visually consistent, and interpretable instruction-based image editing using only open-source components.
RoboLayout is a differentiable 3D scene generation framework that extends LayoutVLM by integrating explicit reachability constraints and a local refinement stage to create semantically coherent, physically feasible indoor environments tailored to the specific capabilities of diverse embodied agents.
The paper introduces Omni-C, a single dense Transformer encoder that compresses heterogeneous modalities (text, audio, and image) into shared representations via unimodal contrastive pretraining, thereby eliminating the parameter overhead and routing complexity of Mixture-of-Expert architectures while achieving comparable performance with significantly reduced memory usage.
This paper proposes a novel multimodal framework, the Clinical-Injection Transformer with a domain-adapted MAE, which integrates routine PAS-stained histopathology images and clinical data to achieve high-accuracy three-class prognosis prediction for pediatric lupus nephritis, addressing previous limitations in data availability and modality integration.
This paper introduces SKETCHGAIT, a robust gait recognition framework that leverages a novel label-free "SKETCH" modality to extract dense structural cues from RGB images, demonstrating that combining this edge-based representation with traditional parsing methods significantly outperforms existing silhouette- and parsing-based approaches.
This paper presents a digital twin-driven V2X trajectory prediction framework for urban intersections that employs a novel twin loss function alongside standard MSE to enforce traffic rules, collision avoidance, and motion diversity, thereby significantly reducing safety violations while maintaining high prediction accuracy and real-time performance.
AutoThinkRAG is a complexity-aware framework for image-text interaction that improves document question answering by routing queries based on difficulty and decoupling visual interpretation from logical reasoning to achieve state-of-the-art performance with reduced inference costs.
This paper introduces Bias-Invariant Subnetwork Extraction (BISE), a method that identifies and isolates fair, bias-agnostic subnetworks within standard pre-trained models through pruning, enabling effective bias mitigation without retraining or additional unbiased data.
This paper introduces "Thinking with Spatial Code," a framework that converts RGB videos into explicit, temporally coherent 3D representations using a specialized spatial encoder and reinforcement learning, enabling large language models to achieve state-of-the-art performance in physical-world visual reasoning on VSI-Bench.
This paper introduces the first coupled robustness verification framework for heatmap-based keypoint detectors that uses a mixed-integer linear program to jointly bound deviations across all keypoints, thereby providing sound and less conservative guarantees than prior decoupled methods.
DreamCAD is a multi-modal generative framework that enables scalable, high-fidelity CAD generation by representing editable BReps as differentiable parametric surfaces for training on unannotated 3D meshes, while also introducing the large-scale CADCap-1M dataset to advance text-to-CAD research.
This paper proposes Adversarial Batch Representation Augmentation (ABRA), a domain generalization framework that synthesizes worst-case bio-batch perturbations via structured uncertainty modeling and angular geometric margins to achieve state-of-the-art batch correction and generalization in high-content cellular screening without relying on additional prior knowledge.
This paper introduces the Post Fusion Stabilizer (PFS), a lightweight, plug-and-play module that enhances the robustness of existing camera-LiDAR fusion 3D detectors against domain shifts and sensor failures by stabilizing bird's-eye view feature statistics and adaptively correcting degraded cues without requiring architectural changes or retraining.
This paper introduces CBM-Suite, a comprehensive framework that addresses key limitations of Concept Bottleneck Models by proposing an entropy-based metric for concept relevance, a non-linear layer to prevent bypassing the bottleneck, and a distillation strategy to close the accuracy gap with opaque models.
This paper introduces interpolated FID (iFID), a novel metric that achieves a strong correlation with diffusion generation FID by interpolating latent representations between dataset samples and their nearest neighbors, thereby overcoming the limitations of traditional reconstruction FID in predicting generative model quality.
This paper introduces Implicit Error Counting (IEC), a reference-free reinforcement learning post-training method that enumerates and weights errors to generate rewards, demonstrating superior performance over Rubrics as Rewards (RaR) in virtual try-on tasks where multiple valid outputs exist and ideal reference answers are unavailable.