@misc{indiciae5efc6557540a,
  title = {Does CLIP's Generalization Performance Mainly Stem from High Train-Test Similarity?},
  author = {Prasanna Mayilvahanan and Thaddäus Wiedemer and Evgenia Rusak and Matthias Bethge and Wieland Brendel},
  year = {2024},
  url = {https://arxiv.org/abs/2310.09562},
  note = {Source identifier: 2310.09562}
}