Monocular Person Localization under Camera Ego-motion

A recent study has highlighted the potential of deep generative models for compressing sensor data in autonomous vehicles, particularly for scenarios requiring remote human assistance. This approach aims to enhance the efficiency of data transmission from sensors like cameras and lidar, which generate vast amounts of information in real-time.

The paper revisits the effectiveness of pre-trained language models (PLMs) in detecting real-world vulnerabilities, highlighting critical challenges such as data leakage and limited scope in existing studies. An extensive evaluation of 18 PLMs on high-quality datasets is conducted, focusing on their performance in vulnerability detection (VD) through fine-tuning and prompt engineering methods.

DeltaDeno introduces a novel zero-shot anomaly generation method that operates without real anomaly samples or training, utilizing a technique called Delta-Denoising to localize and edit defects in images. This method contrasts two diffusion branches driven by minimal prompts, allowing for realistic local defect generation while preserving surrounding context.

A new algorithm named RapidPoseTriangulation has been introduced, enhancing multi-view multi-person whole-body human pose triangulation with remarkable speed and generalization capabilities. This advancement allows for detailed capture of human movements, including facial expressions and finger movements, across various individuals and viewpoints.

Conformal prediction, a model-agnostic approach, is being evaluated for its effectiveness in capturing aleatoric uncertainty, which refers to the inherent ambiguity in datasets due to overlapping classes. This investigation focuses on the correlation between prediction set sizes and the distinct labels assigned by human annotators, aiming to validate the expected performance of conformal predictors in uncertain environments.