All papers

Presents a training-free calibration layer that aligns mismatched cross-modal embedding spaces via local rotation estimation.

Develops a multi-modal image fusion framework that combines structural guidance and interactive gated attention to improve perception.

Presents an automatic scene flow annotation pipeline using articulated mesh binding for non-rigid objects in autonomous driving datasets.

Proposes an attack-defense game framework to reinforce semantic attributes in multi-modal image fusion while preserving visual quality.

Introduces a lightweight velocity corrector to improve prompt following in few-step text-to-image generation models without full fine-tuning.

Proposes a domain adaptive object detection framework using segmentation-guided consistency learning to handle distribution shifts.

Introduces a federated vision-language model tuning framework combining global shared adapters with client-specific personalized projection adapters.

Analyzes what egocentric video embeddings encode and demonstrates complementary sensor fusion with wearable inertial measurement units.

Develops a unified motion deblurring framework using Fourier convolution learning to handle diverse real-world blur kernels.

Demonstrates that visual in-context demonstrations help vision-language models overcome conflicting pretraining priors on counterfactual tasks.

Proposes a training-free token compression method for video language models using video codec motion vectors and residuals as priors.

Investigates internal representation utility in pixel diffusion transformers and proposes Representation Guidance to improve generation performance.

Designs a vision transformer prioritizing feature-axis attention over spatial attention to improve cross-domain transfer in medical imaging.

Stress-tests mechanistic and behavioral claims in a connectome-constrained Drosophila motion-vision neural network model.

Presents a generative video codec supporting both low-delay autoregressive streaming and high-quality bidirectional playback from a single bitstream.

Investigates cross-dataset transfer failures in brain MRI foundation models and attributes them to dataset-specific encoding directions of image content.

Proposes Temporal-Response Guidance to direct latent distillation supervision for video-language models using local temporal perturbation responses.

Develops a training-free framework to refine fine-scale details in generated 3D meshes using focused bounding-box re-encoding.

Proposes a hierarchical model merging method to combine modality-specific trackers into a unified RGB-X tracking framework.

Introduces a 3D vision-language foundation model trained on MRI data with anatomically aligned text and pathology-aware representations.

Presents a training-free framework for prompt-driven audio-video editing that uses flow differences to balance cross-modal changes and preservation.

Introduces PhaseSurf, combining metric-phase fields and volume rendering to reconstruct thin structures and sign-ambiguous geometry.

Accelerates generative MRI reconstruction by learning transitions across discrete k-space intervals combined with stochastic exploration.

Integrates execution progress awareness into vision-language navigation to suppress completed instruction segments and improve multi-step trajectory matching.

Evaluates how scoring configurations and query choices impact the discovery, steerability, and necessity of visual attention heads in VLMs.

Releases an EEG benchmark and evaluates decoding models comparing neural representations during video perception versus internal mental recall.

Guides vision-language navigation training using progress attention and spatial structure losses to improve action grounding and generalization.

Investigates fine-grained visual recognition failures in multimodal LLMs and proposes an inference-time method to mitigate visual cue under-utilization.

Introduces a preview-and-edit framework that diagnoses generation failures from low-cost previews and adjusts cached trajectories during video generation.

Presents a self-supervised interior patch compatibility measure to solve eroded visual jigsaw puzzles without clean boundaries.

Introduces a content-aware 3D Gaussian Splatting framework that adaptively distributes Gaussian primitives based on scene texture and structure.

Presents a training-free generative video super-resolution pipeline combining one-step diffusion and flow interpolation for consumer GPUs.

Proposes a hybrid image tokenizer combining compact 1D global tokens with adaptive sparse 2D tokens for local detail refinement.

Adapts latent video diffusion models with target conditioning and semantic-geometric alignment for coherent generative frame inbetweening.

Studies how text-to-image anchor properties correlate with downstream video generation quality in image-to-video diffusion pipelines.

Presents a reference-free point cloud quality assessment method using successive surface reconstructions to decouple degradation from content complexity.

Combines multimodal graph learning and perspective transformation of visible and infrared feeds for drone-based vessel identification and tracking.

Presents a detection and embedding framework alongside a benchmark dataset for recognizing cartographic signs in historical archival maps.

Develops a distillation framework transferring privileged future trajectory knowledge and safety scoring to improve end-to-end autonomous driving planners.

Audits joint-embedding predictive learning in medical CT scans, showing a bias that degrades performance on unpredictable localized pathology.

Adapts multimodal models for direct continuous bounding box prediction using a lightweight regression head in a single forward pass.

Introduces ContextReID, a framework utilizing co-occurring players and cross-frame sequences to improve player re-identification in broadcast sports video.

Proposes a graph-based token pruning method inspired by sheaf geometry to reduce prefill computation in multimodal large language models.

Introduces Contrastive Vision On-policy Distillation to prevent vision-language models from relying on answer shortcuts rather than visual evidence.

Proposes a contrastive learning framework to help GUI agents distinguish target interface elements from plausible distractor elements.

Introduces an on-policy post-training framework that incorporates visual evidence transfer to improve reasoning in video large language models.

Proposes a quantization module to improve adversarial robustness in self-supervised skeleton-based human action recognition.

Develops a semantic information fusion framework that coordinates distributed edge terminal uploads and slot-based token aggregation.

Investigates decoding register tokens alongside patch tokens in vision transformers to preserve global scene appearance details.

Introduces a streaming causal transformer framework for generating human motion continuously from spatial audio inputs.

Presents parts-specific reweighting and classifier guidance to improve facial details and audio alignment in speech-driven gesture videos.

Proposes a multimodal drag-based image editing framework that generates visual reasoning traces to guide point-based transformations.

Composes internal VAE representations into stage-wise guidance to improve diffusion transformer generative learning without external encoders.

Introduces a reinforcement learning method that decouples sampling from optimization anchors to stabilize video flow generation training.

Analyzes how long skip connections in Diffusion Transformers preserve feature representations for unified visual generation and understanding.

Combines radiomics features, an additive model, and conformal prediction for uncertainty-aware thyroid nodule ultrasound classification.

Introduces a benchmark evaluating text rendering and temporal preservation in text-to-video and reference-to-video generation models.

Introduces an adaptive decision layer and ensemble scheme for chronic stroke lesion segmentation on brain MRI.

Grafts out-of-distribution video exemplars into reward optimization to prevent motion collapse during video generation training.

Proposes a multi-agent evolutionary framework that routes global composition edits and local retouching for constrained infographic generation.