All papers

Proposes a training-free frame packing method enabling vision-language models to process long videos with constant memory consumption.

Introduces masked semantic reconstruction and geometry modulation to better align lifted 2D semantic priors with 3D instance representations.

Develops an interpretable AI framework for breast ultrasound diagnosis incorporating clinical standard descriptors and assessment workflows.

Proposes a prototype-conditioned diffusion method to distill large 3D human motion datasets into compact representative subsets.

Presents a real-time RGB dynamic SLAM framework that removes moving objects at pixel precision using motion discrepancy analysis.

Releases a large gameplay dataset across multiple competitive games to support multi-agent interactive world model research.

Introduces a procedural benchmark and dual-loop evaluation to test whether vision-language models maintain physical and spatial reasoning rules under pressure.

Develops an agent-centric 3D language model that grounds spatial understanding in physical poses without requiring text situation descriptions.

Introduces a training-free open-vocabulary 3D segmentation approach that maintains per-primitive label distributions for clean object boundaries.

Proposes a geometry-grounded video autoencoder incorporating structural depth and geometric constraints to improve latent representations for video generation.

Reformulates graphical interface visual grounding as executable-region mask prediction to pinpoint click actions reliably.

Presents a one-step framework to generate whole-body human-object interactions efficiently using structured latent spaces.

Introduces a six-camera head-mounted system to capture body and bimanual hand motions for embodied demonstrations.

Combines view graph initialization with feed-forward reconstruction using a shared geometric representation to handle uncurated multi-scene image collections.

Enables real-time full-body avatar tracking and personalized appearance on standalone consumer headsets using sparse egocentric sensor inputs.

Presents a depth-guided diffusion method to convert images into seamless repeatable tiles while preserving original local geometry and details.

Uses segmentation guidance to improve density control and loss functions for fine-grained 3D Gaussian Splatting scene reconstruction.

The paper introduces a flow-matching model that jointly identifies garments and estimates full 3D configurations from partial, occluded point clouds.

The paper proposes a learned temporal evidence selection approach to generate grounded text explanations for video anomaly understanding.

The paper presents a rate-adaptive rollout framework that allows interactive video world models to operate consistently across varying temporal rates.

The paper introduces a deep learning framework to estimate building ages from publicly accessible satellite imagery and digital surface models.

The paper presents a training-free semantic feedback framework using vision-language models to iteratively detect and explain video anomalies.

The paper introduces a human-annotated dataset, benchmark, and reward model to improve temporal synchronization in audio-visual generation systems.

The paper introduces a stateful commit-control framework with an evidence blackboard to manage revisions during agentic video reasoning.

The paper analyzes semantic supervision mechanisms in direct speech-to-box grounding models for remote sensing imagery.

The paper presents a phase-aware pose modeling framework and dataset to predict throw distances in elite athletic competitions from broadcast video.

The paper proposes a history-guided pose reselection technique to correct visual localization errors caused by repetitive environmental structures.

The paper introduces an importance-weighted flow matching approach to explicitly learn powered conditional distributions without standard classifier-free guidance.

The paper introduces a benchmark to evaluate direct edits, dependency updates, and visual preservation on technical diagrams, charts, and documents.

The paper proposes a training-free framework that accelerates diffusion inference by forecasting intermediate representations using low-order Bezier trajectories.

The paper presents a dual supportability audit to filter ungrounded cues during forensic distillation for face forgery detection.

The paper proposes a hybrid diffusion method that enforces geometric constraints to generate open-vocabulary 3D scene graphs from point clouds.

The paper proposes a training method that teaches vision-language models to decline to answer when relevant visual evidence is missing.

The paper presents an evaluation framework showing that observation-quality shortcuts can bias motion-based video anomaly detection benchmarks.

The paper proposes a hierarchical alignment method that transfers continuous dynamics and high-level event knowledge into video diffusion models.

The paper analyzes failure modes in unified visual generators using self-reward loops and proposes stabilization methods to prevent quality degradation.

The paper introduces a multimodal entity linking method that combines vision-language models with heterogeneous graph networks for cultural heritage data.

The paper introduces a deep learning framework to rapidly simulate stress fields and mechanical properties of compressed plant fibers.

The paper presents a unified deep learning framework for the automatic detection, classification, and segmentation of lung nodules directly from CT scans.

The paper introduces a monocular vision system to track bronchoscope tip location and map airway surfaces using standard RGB video alone.

The paper presents a benchmark to test how well AI agents can reconstruct dynamic, editable 3D scenes from dynamic reference videos.

The paper proposes a closed-form variational filtering algorithm for the last layer of deep classifiers to improve uncertainty calibration.

The paper introduces a model and dataset for financial candlestick reasoning by generating and executing code to gather visual evidence.

The paper presents a diffusion transformer operating on per-frame velocities to generate rigid point trajectories from a single image.

The paper proposes a neural network trained on synthetic MRI scans to perform cortical surface reconstruction and uncertainty estimation.

The paper introduces a method to detect adversarial chain-of-thought hijacking in backdoored multimodal reasoning models by monitoring reasoning dynamics.

The paper proposes a dual-branch interaction network to model human-robot spatial interactions for predicting human motion during collaboration.

The paper introduces a counterfactual benchmark to evaluate whether vision-language-action driving models truly follow natural language instructions.

The paper presents a framework exploring category, video, and frame prompts to enhance vision-language models for open-vocabulary video recognition.

The paper proposes a framework that evolves a reusable library of skills and tools based on experience for spatial reasoning agents.

The paper presents a method using 3D mesh reconstruction and visibility-aware training to improve cross-view object re-identification.

The paper proposes an adaptive whole-slide image navigation agent that models cross-magnification localization for pathology diagnosis.

The paper introduces a training-free visual token pruning framework using information-density trees to speed up video large language model inference.

The paper introduces a hybrid framework combining explicit depth metricization with data-driven refinement to recover metric depth from relative estimates.

The paper presents a method that fine-tunes image tokenizers on interpolated latent codes to reduce decoding artifacts in diffusion generation.

The paper presents a framework that distills multiple keypoint detectors and dynamically routes them per patch for dense image matching.

The paper proposes a vision-language framework to predict rigid-body connectivity and articulation parameters of objects from single views.

The paper introduces a tool-augmented reinforcement learning framework for evidence-grounded chest X-ray diagnostic reasoning.

The paper proposes a structured pruning method to compress vision foundation models into compact task-specific models under parameter budgets.

The paper presents a dual-branch neural network designed to simultaneously perform pose-invariant face recognition and facial landmark localization.