Table 5.
Optimal CWM parameters and mean-squared prediction errors (MSEs) for estimating human-rated spatial envelope properties from GIST. Note: The cross-validation method for identifying the optimal CWM parameters (spatial resolution and number of clusters) for predicting the spatial envelope properties of Oliva & Ross’ images is described in Oliva & Ross (2010). The model trained on Oliva & Ross’ dataset and tested on SYNS used the optimal model parameters for the training data.
| SYNS/SYNS | Oliva & Ross/Oliva & Ross | Oliva & Ross/SYNS | |||||
|---|---|---|---|---|---|---|---|
| Training/test data | Resolution | Clusters | MSE | Resolution | Clusters | MSE | MSE |
| Mean depth | 1 × 1 | 8 | 0.47 | 4 × 4 | 6 | 0.56 | 1.81 |
| Openness | 4 × 4 | 5 | 0.35 | 8 × 8 | 6 | 0.87 | 1.91 |
| Perspective | 1 × 1 | 5 | 0.99 | 2 × 2 | 4 | 1.95 | 1.30 |