All papers

Proposes a hierarchical multimodal transformer to model long-term behavioral changes across multi-session interactions for cognitive impairment detection.

The paper presents an automated speech recognition and sentence completion system to assist dementia patients speaking Bengali.

Proposes a factor-wise representation framework to selectively treat different distribution shifts in multimodal machine learning.

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

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

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

The paper presents a contrastive learning framework that handles false-negative pairs during zero-shot chest X-ray vision-language pre-training.

The paper introduces an image-conditioned evidence refinement method to improve fine-grained compositional matching in vision-language models.

Introduces a robotic system that converts language instructions and visual inputs into physical light projection actions in spatial environments.

Improves on-policy distillation for diffusion models by restricting feature alignment supervision to deep layers and early denoising steps.

Presents an agentic multimodal framework using a plan-act-observe loop to generate structured, protocol-compliant chest X-ray radiology reports.

Proposes an AI-generated video quality assessment method using visually grounded captions instead of relying solely on source text prompts.

Proposes a training-free closed-form method to refine target object retrieval in 3D Gaussian language maps while preserving open-vocabulary queries.

Distills multi-view visual cues into a latent flow matching network to improve point cloud upsampling at inference time.

Combines visual tracking and vocal fingerprinting to identify and monitor individual dogs across real-world multi-dog video recordings.

Provides region-specific reward modeling to fix incorrect wording and glyph errors in visual text generation models.

The authors introduce CharXiv, a benchmark of realistic scientific charts designed to test descriptive and reasoning capabilities in multimodal language models.

The authors introduce a training loop where multimodal models use text and visual abilities alongside program execution to generate their own training data.

This paper analyzes visual token utilization in high-resolution vision-language models to optimize token pruning across network layers.

This paper presents a mid-training method, corpus, and benchmark to help language models understand and reason about time series data.

The paper proposes an edge-cloud collaborative inspection framework using multimodal LLMs dynamically scheduled based on scene complexity.

This paper introduces an evaluation benchmark using multimodal models to assess generated videos across human preference dimensions.

The paper introduces a decomposed attention mechanism for vision-language models to separate visual and textual token processing.

The paper presents an open-source multimodal language model combining a small language backbone with vision encoders for efficient inference.

The paper introduces a visual and multimodal knowledge graph grounding model designed for visual navigation tasks.

The paper proposes a modular vision-and-language navigation framework combining dynamic memory pruning and language-model spatial reasoning.

The paper presents a multimodal vision-language model for Alzheimer's disease classification and question answering capable of handling missing imaging or clinical data.

Introduces a mixture-of-prompts distillation method to transfer knowledge from hard prompts to soft prompts, improving generalization on unseen classes.

The paper reviews language-vision fusion architectures and deployment strategies for medical diagnosis in low-resource healthcare settings.

The paper presents a framework for balancing multimodal integration using conditional dependency modeling across diverse data signals.

This paper introduces a multimodal meme dataset and a pipeline to extract entities and their semantic relationships.

This paper presents a framework and a dataset to generate holistic full-body co-speech gestures from audio and masked movement priors.

Summary pending (next check).

The paper introduces a method to reduce redundant external knowledge retrieval by estimating what a language model already knows.

Summary pending (next check).

This paper presents a dataset, benchmark, and coarse-to-fine learning scheme to detect tiny oriented objects in imagery.

This paper proposes a tokenized joint source-network coding framework to improve video streaming resilience against network packet loss.

Establishes a training approach to ensure backward compatibility and prevent false positives across versions of neural audio watermarking.

Applies persistent homology to encode global structural connectivity into learned representations of mechanical metamaterials.

Introduces local appearance purification to filter alignment targets and improve representation learning during diffusion model training.

Builds a solar-powered edge vision system to detect and quantify insect defoliation severity in grapevine canopies.

Selects maximally informative pixel tokens directly from vision transformers to train segmentation heads under tight annotation budgets.

Learns robot manipulation priors from egocentric human videos using action prediction rather than rule-based hand retargeting.

Prunes Gaussian occupancy models without preset budgets by identifying and removing Gaussians that do not alter voxel classification decisions.

Adds line-feature tracking to GPU-accelerated visual odometry to improve warehouse robot localization in low-texture environments.

Estimates 3D robot pose from monocular images using graph-guided 2D keypoints and conditional flow matching.

Presents a dataset of 500 robotically imaged cloth materials for training and evaluating neural reflectance reconstruction models.

Introduces a Shapley-based attribution method that jointly explains spatial and spectral channel importance in multispectral imagery.

Separates static and dynamic 3D Gaussians to enable efficient real-time volumetric video streaming on mobile devices.

Accelerates diffusion transformer video generation by increasing attention sparsity and reusing search indices across denoising steps.

Develops a domain adaptation framework to improve pedestrian trajectory prediction under varied social couplings and scene shifts.

Presents an unsupervised deep multiview learning framework to correct systematic multitemporal surface reflectance inconsistencies across satellite observations.

Distills high-fidelity priors into heavily compressed latent spaces to enable faster high-resolution image generation.

Reuses intermediate transformer features across video frames to speed up 3D point cloud semantic segmentation without extra training.

Proposes a local optimization method using Delaunay triangulation to balance geometric consistency with spatial distribution in RANSAC estimation.

Introduces a climate-aware satellite image dataset to evaluate and improve domain adaptive land-cover semantic segmentation across different regions.

Presents a training-free state-correction method to eliminate object residuals during object removal in few-step diffusion models.

Introduces a feed-forward framework with panoramic attention to recover geometrically consistent 3D point clouds from a single panorama.

Improves dynamic 3D Gaussian splatting for fast-moving objects by injecting seeds in under-reconstructed dynamic areas.

Reduces the inference computational cost of vision graph neural networks by applying graph coarsening via MinCUT pooling.