Hierarchical abstraction drives human-like 3-D shape processing in deep learning models

Fig 13

Correlation between model and human accuracy patterns across all experiment conditions.

Models with downsampling show significantly stronger alignment with human.