Figure 5.

Response normalization for object pairs along continuous shape dimensions. A, B, Individual examples of tuning curves obtained for two neurons recorded in experiment 2 (A, same neuron as in Figs. 2C–E and 3B, left; B, same neuron as in Fig. 3B, right). The abscissa is the shape distance (i.e., shape dissimilarity) within the tested morph line (shapes corresponding to some of the tested distances are shown below each shape axis). The origin of the shape axis is the preferred shape of the neuron obtained from the recording screening procedure (see Materials and Methods). The gray patch shows the region of shape space initially tested to obtain the preferred shape (unit shape distance; see Materials and Methods). The horizontal dotted line indicates the background rate of the neuron. Morphed shapes were sampled within either the unit distance (A) or a larger shape range (B) that included a stimulus drawn from a different shape space (data points at the far right in B). For both neurons, responses to the object pairs (black line) are very close to the average (green line) of the responses to the constituent objects of the pairs presented in isolation, i.e., to the average of the horizontal dashed line and the red line (open circles). Error bars are SE of the mean firing rate. C, Population average of 26 tuning curves obtained from the 15 most selective neurons recorded in experiment 2 for single and pair object conditions (see Results). These tuning curves were background subtracted, aligned to the preferred object (0 on the abscissa), and normalized by the response to the preferred object. The inset shows these 26 normalized tuning curves for responses to single objects (thin gray lines) and their average (thick red line). The red line (open circles) in the main panel shows the population average of the responses to single objects and included single object conditions outside the unit shape distance (gray patch). The horizontal dashed line shows the population average of the responses to the preferred object of each neuron (i.e., the shape at value 0 on the abscissa). The black line shows the population average of the responses to object pairs containing both the preferred object and another object sampled along the abscissa. The green line shows the average model prediction (population normalized average of average model curves as in A and B). The cyan line shows the prediction of the CCI model (see Results). Error bars are SE of the population averages. Although different morph lines were tested for different neurons, example shapes are shown below the abscissa from a representative morph line. The dotted line is the background rate. Only 11 of 15 neurons (for a total of 18 of 26 responses) were tested outside the unit distance (gray patch) and contribute to the points outside this range. This plot is nearly identical when constructed from conditions in which the preferred object was either in the best or second best RF position (top or bottom; data not shown), i.e., forcing every neuron to contribute only one tuning curve to the population average.