Robustmixgen: data augmentation for enhancing robustness of visual-language models in the presence of distribution shift

HIGHLIGHTS

• What: The authors provide an introduction to techniques to a single modality. Building Robustness is defined as the extent to which a model is upon these prior investigations, the experiment aims to perform influenced by circumstances such as data loss, data con- a comprehensive analysis of the limitations of MixGen , tamination, and jitter . In this experiment, the has been limited exploration into how data perturbations can object selected for removal from the image is determined as the object with the largest bounding box area, as indicated by impact model performance . In this experiment, the authors examine the . . .
  

   

  

  If you want to have access to all the content you need to log in!

  Thanks :)

  Username or E-mail:

  If you don't have an account, you can create one here.