All papers

The paper introduces a dataset and synthetic generation pipeline focused on recognizing accessibility-critical objects to assist blind navigation.

The paper proposes a reinforcement learning framework using quality-aware rewards to guide multimodal knowledge graph construction for retrieval-augmented generation.

The paper investigates novelty and anomaly detection techniques across rain gauge sensor networks and computer vision scenarios.

The paper benchmarks how detector architectures, model scale, and crowd density influence multi-pedestrian tracking performance in video sequences.

The paper proposes a hierarchical two-loop optimization framework that uses residual errors to trigger active evidence acquisition during inference.

The paper introduces topological regularization to align latent spaces across independently trained autoencoders without inter-model communication.

The paper proposes compressed window autoregression to prevent repetitive bounding box errors in vision-language object grounding tasks.

The paper proposes a deep unfolding network integrating manifold optimization for unsupervised feature selection in image processing.

The paper proposes a depth-conditioned diffusion transformer architecture to suppress overlapping bone structures in chest X-ray images.

The paper introduces an embedding model that jointly encodes vision, language, and action to retrieve relevant historical trajectories for robot learning.

The paper introduces a generative pipeline to reconstruct multi-room 3D environments from sparse panorama images using geometric optimization.

The paper introduces a synthetic world generator and benchmark for evaluating 3D hyperspectral material unmixing within novel view rendering.

The paper introduces a rank-calibration framework to improve proposal localization quality for temporal action detection in long untrimmed videos.

The paper proposes a loss function that reduces spatial neighborhood discrepancies in space-filling curve representations for 3D point cloud segmentation.

The paper proposes an optimization method for 3D Gaussian Splatting that improves rate-distortion performance when compressing scenes with image codecs.

The paper introduces TEXTVERIFIER, a verification framework using character structural sequences to detect fine rendering defects in generated Chinese text.

The paper introduces FLARE, a visual speech framework combining cross-modal distillation from audio features with dual-stream face and lip modeling.

The paper proposes a step-wise alignment method to suppress harmful content in text-to-image models while maintaining generation semantics.

Uses factorial interventions to locate failure points in multi-image vision-language reasoning pipelines.

Introduces a benchmark dataset and baseline architecture for reconstructing standalone 3D garments from single images.

Adapts both field of view and fixation counts per image to optimize computation in active visual recognition.

Proves theoretically that self-conditioning enables genuine multi-step reasoning in flow matching where standard samplers fail.

Delineates Arctic coastal polynyas from satellite sea ice concentration data using hierarchical partitioning and region growing.

Transfers hyperspectral tokens into frozen vision-language-action models via learned continuous residuals without retraining.

Equips frozen 3D geometry foundation models with an associative external memory to revise past predictions across long horizons.

Reconstructs mixed-reflectance 3D surfaces by integrating microfacet normal consistency into neural implicit distance fields.

Adds content-addressable associative memory to latent world models to prevent collapse and improve long-horizon planning.

Couples a persistent symbolic world engine with an egocentric generative model for continuous interactive audiovisual simulation.

Identifies scale mismatches as the cause of trajectory drift in monocular human recovery under rapid camera movement.

Presents a trust-region optimization method using dual quaternions to estimate 3D object pose from 2D correspondences.

Incorporates hyperspectral imaging descriptors into 3D Gaussian Splatting loop closure to prevent visual aliasing in planetary SLAM.

Extracts generative image priors directly from the loss-gradient geometry of pretrained discriminative models without retraining.

Audits physics-guided underwater image restoration models, showing image-formation priors provide minimal benefit under full training data.

Generates quad-only 3D meshes autoregressively using a singularity-aware tokenization scheme that preserves strict topology.

Formulates computational compressive imaging reconstruction as an Ising model solved via simulated annealing.

Introduces a benchmark dataset to localize impact moments in drone first-person-view crash videos for post-flight analysis.

Improves active domain adaptation for difficult classes using region-based positive manifold coverage and uncertainty scoring.

Introduces a diagnostic question-answering benchmark to evaluate whether text-to-video models follow real-world physical laws.

Evaluates four world-model integration paradigms to identify what improves vision-language-action policies in robotic tasks.

Combines 3D Gaussian Splatting with ray-traced occlusion estimation to enable relighting and material decomposition.

Recovers full-body human motion from flexible sensor subsets using a cross-modal masking mechanism and single-step flow matching.

Integrates video codec dynamics with visual features to improve unsupervised temporal action localization without manual annotations.

Combines an adapted editing model with an agent harness to perform instruction-driven four-kilopixel image editing.

Analyzes why feature-based test-time adaptation fails to recover accuracy on corrupted inputs in generative multimodal language models.

Develops a closed-loop distillation pipeline where student feedback refines teacher models to improve adversarial robustness.

Applies optimal transport metrics to quantify graded annotation agreement and label discrepancies in object detection datasets.

Trains a reinforcement learning controller to manage frame memory dynamically for visual object tracking in foundation models.

Generates a filtered synthetic dataset of image-based instruction-response pairs for training multimodal reasoning models.

Analyzes reconstruction failures on under-ice underwater vehicle video and introduces a tool to flag camera baseline errors.

Adapts a frozen segmentation model using a dual-branch adapter to improve cell nuclei segmentation across varied histopathology domains.

Constructs a digital twin from computed tomography scans to plan accurate surgical trajectories for ventricular shunt insertion.

Introduces a hierarchical multi-task benchmark to evaluate fine-grained visual world knowledge in multimodal large language models.

Addresses cyclic preference graphs in reinforcement learning for video generation models using an anchored preference walk.

Models entity persistence and relation transitions separately to improve video scene graph generation accuracy during dynamic scene changes.

Uses representation-space supervision to initialize autoregressive video distillation students for faster and clearer generation.

Transfers capabilities from text-to-video foundation models to a unidirectional streaming model for real-time video object removal.

Introduces dataset-level regularization stored in memory to prevent representation collapse in self-supervised learning with small batch sizes.

Learns continuous latent actions from video data using an orthogonal dictionary to build cross-domain world models.

Applies low-rank continual pretraining to vision transformers to improve task alignment in prostate magnetic resonance imaging segmentation.

Distills bidirectional video diffusion models into causal autoregressive models using an efficient single-sample supervision strategy.