Fig. 7.

Segmentation can be decoded with higher accuracy from segmentation patches. We used a linear classifier to discriminate between four classes (figure, background, top edge, bottom edge; chance performance: 25%). For each data point, we performed 100 iterations of randomly selecting a given number of neurons (X axis) from all cells recorded from segmentation patches (blue), globs (red), or control regions (yellow) and randomly selecting one half of the trials for training the classifier and the other half for testing to determine cross-validated decoding accuracies (Y axis). Firing rates averaged across 0 ms to 250 ms after stimulus onset for each neuron were used as features for the classifier. Solid lines indicate cross-validated decoding accuracies averaged across iterations. Dashed lines indicate the 95% CI across the 100 iterations.