Latest Articles for Computer Vision

A new approach improves AI's ability to handle unusual data.

A new training strategy improves 3D vision systems’ resistance to misleading inputs.

LLaVA-3D combines 2D and 3D insights for deeper spatial reasoning.

Exploring the use of synthetic data to enhance DRL in real-world applications.

InterNet enhances homography estimation by learning from images without labeled data.

Learn about image denoising techniques to improve clarity and quality.

A fresh dataset addresses viewpoint shifts in depth estimation for autonomous driving.

A method that combines event data and traditional frames for better motion analysis.

A new approach enhances the learning process between teacher and student models.

A new method to balance general knowledge and task-specific adaptation in models.

AP-VLM boosts robot perception and interaction through active perception techniques.

P4Q combines fine-tuning and quantization for efficient visual-language model performance.

Introducing TA-Cleaner, a method to improve multimodal model defenses against data poisoning.

A new framework for lightweight and effective visual object tracking.

CAMOT improves multi-object tracking by estimating camera angles and depths.

SimVG improves visual grounding by linking text to specific image areas more effectively.

EAGLE model and dataset enhance understanding of egocentric videos.

New method improves crowd counting accuracy and model reliability.

Examining how SSL models memorize data points and its implications.

New methods improve efficiency and accuracy in SSM-based vision models.

A new method improves 3D shape accuracy in dynamic scenes.

New methods improve speed and quality in image deblurring tasks.

A new method improves knowledge transfer in machine learning models.

Introducing a method for AI to generate images without large labeled datasets.

GeCo improves object counting with fewer examples, enhancing accuracy and reliability.

CION advances person re-identification by focusing on identity correlations across videos.

A new method improves gaze target detection with less labeled data.

A new framework improves pixel labeling by addressing uncertainty in semantic segmentation.

This study assesses the effectiveness of pre-trained models in Earth Observation applications.

A new method improves data alignment, especially with noisy datasets.

A look into how CNNs learn image features and their universal similarities.

Exploring methods to improve multimodal models in breaking down visual questions.

TrojVLM exposes vulnerabilities in Vision Language Models to backdoor attacks.

A new framework improves data generation across multiple sources using energy-based models.

SATA improves the robustness and efficiency of Vision Transformers for image classification tasks.

A new method improves object recognition using masks without detailed labels.

PPLNs enhance event camera data processing for improved machine vision capabilities.

Analyzing the effects of pruning methods on GoogLeNet's performance and interpretability.

Innovative methods for enhancing depth maps vital for augmented and virtual reality.

A method to enhance model performance despite incorrect data labels.