All papers

This paper proposes a parameter-efficient framework using frequency-routed adaptation to translate thermal face images into visible-spectrum photographs.

This paper introduces a vision-language model trained on paired microscopy image-text data for open-set biological image classification and retrieval.

This paper introduces representation supervision to enable effective test-time recurrent depth scaling in diffusion Transformers.

This paper introduces a grounded alignment framework using anomaly detection datasets and patch-level policy optimization to fix distortions in flow generators.

This paper proposes a multi-stage cascade with heterogeneous mixture-of-experts lesion segmentation for diabetic retinopathy screening and grading.

This paper introduces a physics-guided uncertainty model to improve dynamic 3D Gaussian splatting in turbid underwater environments.

This paper presents a lightweight, interpretable neural network architecture for resource-constrained time-series classification on microcontrollers.

This paper presents a diffusion MRI foundation model pretrained on masked reconstruction to enable diverse downstream neuroimaging tasks from sparse scans.

This paper introduces a tail-aware evaluation method that estimates web-scale image retrieval performance using small random distractor samples.

This paper introduces a multimodal language-driven framework combining polarization and RGB cues to identify degradations and restore underwater images.

This paper proposes an uncertainty-aware semi-supervised network with supervised contrastive learning to detect sugarcane leaf diseases from limited labeled data.

This paper presents a neural surface reconstruction formulation from range scans using sensor visibility constraints and robust ray intersection regression.

This paper proposes a parameter-free training module that dynamically reshapes prediction errors to improve low-light image enhancement models.

This paper introduces a rendering-free closed-loop simulation method that predicts internal latent states to evaluate autonomous driving systems without image generation.

This paper introduces a data-aware subspace adaptation method using semantic proxy sets to improve generalizable AI-generated image detection.

This paper proposes a rectified generative replay framework enabling robotic world action models to continually learn new tasks without storing past data.

This paper presents a foundation model and a hundred-million-image dataset for mask-based object, text, and watermark erasure in images.

This paper proposes a multimodal variational autoencoder framework that separates task-relevant information to improve clinical predictions even with missing modalities.

This paper presents an asymmetric dual-student framework for semi-supervised medical image segmentation using adaptive reinforcement-guided augmentation.

Proposes FlowCred, a hierarchical credit assignment framework to coordinate prompt optimization and video generator post-training.

Benchmarks multimodal large language models paired with object detectors for zero-shot human-object interaction recognition.

Introduces VARI, a group-structured intent alignment framework to improve text-to-video generation robustness across equivalent prompts.

Proposes TRACE, a training-free transductive inference framework to reduce the prototype estimation gap in few-shot action recognition.

Surveys traditional model-based, signal processing, and machine learning methods for infrared small target detection.

Introduces PixText, a pixel-space grouped generation model for efficient localized text correction in generated images.

Proposes LongForcing, a temporal manifold matching distillation approach that scales long-horizon video rollout stability.

Surveys deep learning approaches, models, and critical scenarios for infrared small target detection.

Presents Omni-RS, a unified framework and dataset supporting multiple remote sensing image understanding and generation tasks.

Introduces CityCode, an agent-operable representation generating parameterized 3D urban environments from sparse low-altitude aerial images.

Presents a visual servoing approach using binormalized Plucker line coordinates to decouple rotation and translation control.

Presents FaithRS, a self-reflection framework with a fine-grained dataset to detect and correct reasoning hallucinations in remote sensing models.

Introduces SchemaPose, a schema-guided framework for RGB-based category-level 9D object pose estimation.

Proposes Noise2Match, a self-supervised low-dose CT denoising method that aligns adjacent slices using adaptive non-rigid matching.

Presents ReDiS, a self-evolution framework separating exploration and candidate diagnosis to improve multimodal reasoning in remote sensing.

Introduces Bifocal, a training-free mixed-granularity visual token pruning method for multimodal models to protect fine-grained task accuracy.

Proposes CoTL, a method to learn compact latent tokens from unposed multi-view images to reconstruct 3D Gaussian scene representations.

Introduces EmoGenie, a benchmark evaluating expressed and evoked emotions in text-to-image generation and multimodal language model understanding.

Proposes a self-supervised plug-and-play restoration framework that learns score priors directly from intermediate states for scientific imaging.

Introduces MFVSR, a multi-style benchmark evaluating localized errors, temporal stability, and reliability in video subtitle removal.

Proposes Simplified Matching Distillation, a regression-based on-policy method that enables four-step generation for diffusion models.

Detects talking-head deepfakes in dynamic videos by matching speaker distance estimates with distance-dependent acoustic energy variations.

Presents STCNet, a dual-domain framework using spatial-temporal consistency learning to reconstruct sparse-view 4D CBCT images with fewer projections.

Evaluates video benchmarks, showing that unexamined setup choices shift accuracy significantly and proposing a standardized evaluation framework.

This paper evaluates whether object pointers from the Segment Anything Model can transfer to downstream clinical tumor prediction tasks.

The paper proposes an unbalanced optimal transport method to reduce minibatch fluctuations in teacher-free one-step generative models.

This paper proposes a lightweight self-supervised neural network trained without clean reference images to reduce noise in chest radiographs.

The paper proposes a benchmark and an uncertainty calibration method for aerial map matching in navigation systems.

This paper introduces parameter-efficient shared adapters to combine multiple MRI modalities for brain tumor segmentation with lower complexity.

The paper proposes measuring prediction output agreement between full-precision and quantized models to identify unreliable classifications without ground-truth labels.

This paper presents a framework adapting diffusion transformers into deterministic, one-step decoders for extreme image compression.

The paper adapts multimodal language models using scoring heads and computation caching for fast video decision-making and interaction.

This paper introduces a multi-sensor, multi-temporal satellite imagery dataset designed for pretraining Earth observation foundation models.

The paper compares spatial feature grids and tracked node coordinates for maintaining object memory under partial robot observability.

This paper studies texture interpolation using appearance-aware texton matching instead of standard positional coordinate blending.

The paper proposes an agent framework that plans multi-slide exploration and visual tool use for patient-level pathology visual question answering.

This research statement outlines a conceptual framework and future directions for robot scene understanding under incomplete sensory and environmental data.

The paper introduces a reciprocal self-improvement framework to balance agent understanding and image generation roles within multimodal models.

This paper applies semidefinite programming relaxation to estimate orientations of symmetric molecular structures from cryo-electron microscopy images.

The paper introduces a synthetic-to-real deep learning framework for directly tracing filament instances in cryo-electron microscopy images.

This paper presents an interface framework applying visual transformations and adversarial attacks to create assignment text resistant to machine extraction.