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

Figure 8.

Figure 8.

Percentage point drop in the classification performance on CIFAR10 (top) and SVHN (bottom) as the Gaussian noise severity increases. The percentage drop is calculated with respect to the classification performance of the DenseNet model in the absence of any Gaussian noise. Without noise, the DenseNet classification performance for 500 Test Samples and All Test Samples for CIFAR10 is 82.93% and 83.80%, and for SVHN is 96.06% and 95.65%. While the performance of all models drop as degradation increases, the drop of topological fusion models is less compared to just the DenseNet model. Note, the y-axis is scaled different for each dataset.