Figure 2

Global mean absolute error (MAEpoint-wise) of 10-2 visual field prediction from Inception-ResNet-V2 and Inception V3. The prediction errors of Inception-ResNet-V2 were significantly lower than those of Inception V3 regardless of input images. mGC/IPLT macular ganglion cell/inner plexiform layer thickness; RNLFT retinal nerve fiber layer thickness; InResNetV2 Inception-ResNet-V2.