arXiv Computer Vision research summaries for April 01, 2024.

Today's Research Themes (AI-Generated):

• New open-vocabulary SGG framework leverages pre-trained vision-language models for enhanced scene graph generation

• Image-Conditioned Caption Correction task improves zero-shot performance of generative vision-language models without labeled data

• LLaMA-Excitor introduces a lightweight module to stimulate LLM potential while preserving pre-trained knowledge

• TEAR method significantly enhances the scalability and efficiency of outlier-robust 3D registration

• GyroDeblurNet leverages gyro sensor data for state-of-the-art single image deblurring performance