2 papers across 2 sessions
object-centric inductive biases to a address the binding problem in CLIP-like pretraining