Figure 7.

ConvCNPs-Net for MNIST with MR. (a–d) show that the missing rate takes values from [0.1, 0.3, 0.7, 0.9], respectively, to test the performance of the trained ConvCNPs-Net and ConvCNPs–SDE models. For each panel, the top row shows the original ID data MNIST, the second row demonstrates the mask based on MR, the third row exhibits the ID MNIST data with the mask, and the bottom row displays how the masked MNIST data are completed by the ConvCNPs of the ConvCNP–SDE model.