All papers

Introduces PEBench, a benchmark and evidence-grounded framework to directly evaluate prompt enhancers for video generation without rendering full videos.

The paper proposes a streaming video super-resolution framework reusing intermediate latent diffusion features across frames via cross-step attention.

The paper proposes predicting pretrained DINO representations directly to guide flow matching models, avoiding a secondary ODE denoising process.

The paper introduces Memory Forcing, partitioning cache into archive and working banks to prevent mid-horizon history loss in streaming video generation.

The paper introduces Sora100K, a dataset structuring AI-native video creation workflows into multi-turn creation trajectories and editing operations.

The paper introduces Skill-V, a verifiable agent skill library that treats skills as falsifiable contracts tested against environment execution outcomes.

The paper proposes PAVG, a dual-branch video generator modeling distinct solid-gas dynamics and cross-attention interactions for physically grounded motion.

The paper proposes Vantage, a training-free framework pairing VLMs with 3D models to plan and synthesize relevant viewpoints for spatial reasoning.

The paper proposes a complex-valued neural network to fuse audio and visual similarity streams for open-vocabulary audio-visual event localization.

The paper proposes VEDJE, compressing video features into reusable caches with feature-change prediction for efficient and scalable video-text retrieval.

The paper introduces FreeInpaint, a feed-forward framework performing 3D scene inpainting directly from unposed multi-view images using masked attention propagation.

The paper proposes Relative Patch Response Learning to improve AI-generated image detection by mitigating self-attention-induced contextual shifts across mixed patches.

The paper proposes MIDAPN, a unified spatiotemporal network using identity-aware graphs and temporal learning across numerical and narrative-flow time series.

The paper evaluates whether contrastive encoder adversarial alignment metrics transfer to target recovery by independent models across different perturbation budgets.

The paper introduces LGWM, an action-conditioned world model predicting next-screen latent representations to verify GUI agent state transitions.

The paper uses activation steering of affective circuits in vision-language models to increase hazard-detection recall for post-disaster damage assessment.

The paper proposes EUA-Cal, regularizing late training against early-stage predictive uncertainty to prevent models from developing overconfidence.

The paper introduces ViMoD, maintaining a compact visual context while dynamically routing back to fine-grained tokens during multi-step reasoning.

The paper presents LIVIN, a spatial and embodied intelligence benchmark featuring digital twins of 30 lived-in homes with human-reviewed object arrangements.

The paper introduces benchmarks and summarizes winning solutions for city-scale audio-visual reasoning and spatial intelligence at the Perception Test 2026.

The paper presents DVLA-RL++, combining complementary semantic purification and reinforcement learning gating to reduce context contamination in few-shot learning.

The paper proposes data-space iteration, a few-step generative method removing flow discretization by directly refining noise predictions in data space.

Recasts referring expression segmentation as a closed-loop policy iteratively refining explicit boundary contours via visuomotor actions.

Presents an agent harness that pairs frozen multimodal language models with modular navigation skills for general robot navigation tasks.

Uses diffusion inpainting to erase lesions from mammograms, creating synthetic healthy counterfactuals to augment breast cancer classification training.

Proposes a training framework reconnecting backward gradient paths across autoregressive chunks to improve long video generation consistency.

Introduces an end-to-end diffusion model operating directly in radar space for few-step probabilistic precipitation nowcasting.

Presents a semi-automatic stride-independent image patching method guided by user input to train deep learning segmentation models.

Proposes a set-associative spatial memory retrofit for streaming 3D reconstruction to preserve distinct viewpoints at shared spatial locations.

Investigates extracting spatial, attribute, and relational modifier grounding directly from frozen multimodal large language model representations using optimal transport.

Introduces an adaptive tokenization method allocating variable compute budgets to image diffusion based on visual complexity and spatial layout.

Develops a framework incorporating character animation principles like exaggeration into generative motion models via adaptation and guidance.

Introduces a multi-dimensional benchmark evaluating reasoning-informed visual editing across temporal, causal, spatial, and logical operations.

Proposes representing robot policies entirely in deterministic stateful code instead of querying vision-language models during run-time execution loops.

Presents a framework linking game engines with a neural renderer to simulate interactive virtual environments for training embodied agents.

Proposes a decoupled 4D reward framework separating static scenes from dynamic objects to improve consistency in generative video models.

Presents a test-time alignment framework that grounds generative 3D foundation models in observed geometry and appearance without model retraining.

Proposes a distributed multiplayer world model architecture where individual clients generate synchronized video views by exchanging player state fields.

Introduces a unified multimodal agent using evidence graphs to supervise external retrieval and reasoning across images and videos.

Presents a dense correspondence framework combining generative and semantic representations to handle non-rigid image editing and generation.

Introduces a benchmark evaluating metric point-to-point distance estimation in 3D foundation models using physical ground-truth measurements.

Introduces a lifting-free transformer model that directly synthesizes egocentric video from exocentric recordings without explicit 3D reconstruction.

Designs a vision transformer using a single recurrent block with depth-programmed experts to reduce stored parameters without accuracy loss.

Presents a previsualization workflow combining physical video capture and generative reshooting to evaluate and manage visual hallucination risk.

Introduces an audio-visual caption benchmark using atomic evaluation units to evaluate multimodal large language models reliably.

Presents a closed-loop goal-directed video world model that iteratively plans and realizes atomic steps for procedural tasks.

Generates seamlessly looping 3D cinemagraphs from static 3D Gaussian splats using periodic deformation fields guided by video models.

Studies how scaling to 30,000 hours of egocentric video affects world model fidelity for agents versus object interactions.

Proposes a self-evolving image editing method using rubric-guided contrastive verification without external reward models or human annotations.

Introduces an adaptive patch-wise denoising framework that stops easy patches early to speed up autoregressive video diffusion.

Benchmarks vision-language models on the Stroop paradigm to evaluate feature competition and text-versus-color bias.

Evaluates explainability methods across CNN, ViT, and Mamba models using eye-tracking gaze data as ground truth.

Proposes a hybrid linear and sparse attention framework to accelerate feed-forward 3D foundation models across many views.

Presents an automatic reference-free method to reconstruct and estimate food portion volumes using egocentric stereo video.

Develops a kernelized parameter sensitivity atlas to amortize input-specific gradient attribution across vision, forecasting, and biology models.

Introduces a shift-aware identity-guided latent refinement framework using dual tokens to resolve target drift in referring audio-visual segmentation.

Presents a training-free framework combining variance-adaptive particle search with Metropolis-Hastings refinement for budget-constrained audio-visual alignment.

Proposes a structure-aware matching framework exploiting directional and low-frequency patterns for UAV thermal-to-satellite geo-localization.

Introduces an asynchronous prefix denoising framework for extreme latent compression in diffusion models.

Presents a structured visual reasoning agent that uses intermediate symbolic representations to solve complex visual CAPTCHAs.