All papers

The paper introduces a compact vision-language-action model using register tokens as persistent memory to reduce control oscillations in robotics.

The paper proposes a training-free method combining global and regional representations across encoder layers for few-shot vision-language adaptation.

The paper introduces an open-source framework and codebase for building generalist computer-use agents that operate across visual and code interfaces.

The paper presents a refiner that models motion proposals as heterogeneous temporal graphs to improve trajectory forecasting for autonomous vehicles.

The paper investigates learning physical damping properties from impact audio and distills this knowledge into vision models.

The paper uses wavelet subband decomposition with language models to improve entropy modeling for lossless image compression.

The paper proposes a training method combining data refinement with reinforcement learning and distillation to improve long-horizon multimodal search agents.

Analyzes image denoisers on repetitive textures and finds fine-tuning on textures leads to catastrophic forgetting of generic images.

Presents an annotation-free retinal vessel segmentation pipeline combining classical filtering, Bayesian deep learning, and graph reconstruction.

Introduces a reinforcement learning objective that groups prompt tokens into semantic spans to refine text-to-image prompt rewriting.

Maximizes batch Gram matrix determinants during initial diffusion noise optimization to improve overall sample diversity in image generation.

Combines vision transformer features with color histograms to cluster public transit vehicles in surveillance video without manual labels.

Frames spatial understanding tasks as candidate classification using pretrained attention matching to enable low-latency spatial decisions.

Proposes an online reinforcement learning approach that uses attention routing as stochastic exploration to align visual diffusion generators with rewards.

Introduces a benchmark and method to detect, describe, and recognize activities for every person visible in complex multi-person video scenes.

Presents a real-time tracking-by-detection method that dynamically scales search regions based on predictive motion uncertainty.

Introduces a geometry-based attention router that selects visual memory contexts to speed up chunked video world model generation.

Presents a training-free framework that balances identity consistency and narrative scene adaptation for multi-frame visual storytelling.

Introduces a feed-forward model that predicts dense 4D point correspondences across articulated human bodies from monocular RGB images.

Accelerates autoregressive video generation by verifying and freezing intermediate tokens early to reduce redundant computation during denoising.

Proposes a framework to dress 3D meshes with aligned materials by decoupling visible surface estimation from observation-guided 3D completion.

Presents a dual-kernel attention mechanism that uses 3D geometric positions to improve spatial reasoning in vision-language models.

Introduces a benchmark spanning ten tasks to evaluate multimodal model verification and rejection explanations on generated 3D content.

Develops a unified world action model combining 3D perception, visual future prediction, and goal-directed planning for egocentric dexterous manipulation.

Introduces an adaptive step-size caching method for diffusion transformers to speed up generation by reusing intermediate network evaluations.

Proposes a training-free token merging method that leverages diffusion dynamics to accelerate image generation while retaining background details.

Repurposes a pretrained video generator into a single-pass feed-forward system to estimate 4D motion, depth, and camera poses.

Presents a benchmark with human-annotated visual evidence to evaluate retrieval and question answering across long documents.

The paper introduces ZoomCap, an adaptive visual zooming and evidence-routing framework to improve fine-grained image captioning supervision.

Identifies a specific vision transformer attention head that spuriously attends to medical image corner text instead of lung pathology.

Proposes a reinforcement learning credit-assignment method for document parsing models to align rewards with structured page elements.

Proposes an event-centric framework to align audio and video features when generating detailed video descriptions with language models.

The paper introduces a hierarchical benchmark for evaluating visual generative models on multi-step world state prediction tasks.

The paper presents a metric and benchmark measuring how quickly environment dynamics can change before agent decision-making fails.

The paper introduces a reasoning-augmented reconstructive post-training framework that unifies multimodal understanding and image generation tasks.

The paper introduces an inference-time guidance method that enforces temporal logic constraints during text-to-video generation.

The paper analyzes how vision-language models process negation by comparing model attention mechanisms with human eye-tracking data.

The paper presents a self-play reinforcement learning framework to simulate interactions between vehicles, pedestrians, and cyclists without human demonstrations.

The paper presents a zero-shot vision-language method for generating task-oriented robotic grasps in 6 degrees of freedom.

The paper proposes a self-play framework to train multimodal foundation models using text proxies instead of paired cross-modal datasets.

The paper presents a hierarchical reinforcement learning framework for non-prehensile 3D manipulation of multi-object stacks using quadruped robots.

The paper uses latent eigenvectors from CLIP to select smaller, representative training datasets for diffusion models.

Introduces an equivariant self-supervised learning framework that avoids topological collapse in 3D point cloud masked autoencoding.

Introduces a multimodal harness and benchmark pairing continuous video with clinical records to assist patient monitoring in hospitals.

The paper presents an agent system that generates, edits, and renders interactive virtual worlds from text using executable code rules.

The paper examines how increasing visual token budgets and cropping key evidence improves document retrieval-augmented generation in vision-language models.

The paper presents an agent orchestration framework that trains smaller language models to navigate, inspect, and reason over long multimodal documents.

The paper introduces an in-context memory method storing latent transitions between visual observations to assist robot manipulation without extra annotation.

Presents methods for query-guided video grounding across object localization, relationship detection, and temporal moment identification.

Proposes a training framework enabling aerial vision-language agents to navigate in full action chunks and stop reliably at destinations.

Presents a full-duplex conversational model that generates real-time speech alongside synchronized facial expressions and body gestures.

Investigates whether training multimodal models on visual generation tasks transfers accessible factual knowledge into text question answering.

Surveys the application of artificial intelligence methods to humanitarian disaster response and displacement monitoring.

The authors propose an indoor navigation framework that uses user-provided floor plans and topological memory without prior robot site maps.

Develops a domain harness and policy learning pipeline to improve language model agent execution in computer-aided design environments.

Combines multi-layer vision transformer representations with a fusion decoder to localize pixel-level generative image inpainting manipulations.

Introduces a causal-beam transfer framework that extracts relevant subtask representations from human videos to guide robotic manipulation.

Presents a unified framework with synthetic data generation and benchmarks to train models on embodied spatial reasoning and actions.

Introduces a framework that aligns interaction regions across objects to improve 3D affordance grounding on unseen objects.

Investigates how user trust influences delegation and verification choices during human interaction with multimodal video question-answering systems.