Skip to main content
. Author manuscript; available in PMC: 2020 Sep 28.
Published in final edited form as: Conf Comput Vis Pattern Recognit Workshops. 2020 Jul 28;2020:3639–3648. doi: 10.1109/cvprw50498.2020.00425

Table 3.

Image classification accuracy results for CIFAR10 and SVHN datasets, with the mean ± std values calculated over three runs. P-value is calculated with respect to the base DenseNet model.

Method CIFAR10 SVHN
Mean±SD p-Value Mean±SD p-Value
DenseNet 83.80±0.12 - 95.65±0.00 -
DenseNet + PI 84.37±0.21 0.0153 95.86±0.01 <0.0001
DenseNet + Image PI-Net 84.82±0.19 0.0160 95.95±0.08 0.0038
DenseNet + Image PI-Net FA 84.69±0.38 0.0195 95.84±0.06 0.0063
DenseNet + Image PI-Net FS 84.59±0.17 0.0032 95.95±0.07 0.0020