Visual perception of liquids: Insights from deep neural networks
Fig 8
The Spearman correlations for each layer and predictor.
We included two final layers, one with the original 4096 units and one where the last layer was retrained to contain only 15 units (see second half Results: Layer activations). Only the B-LAB predictor for layer ReLU 4–15 was not significant (p > 0.05).