Figure 2. Overview of the pipeline of EMGauss. Given a vol-
ume EM dataset (top), the gray-shaded regions denote the miss-
ing intermediate slices to be interpolated, while the visible regions
represent the available anisotropic training data. We interpret the
low-resolution z -axis as a temporal dimension and employ a de-
formable 2D Gaussian splatting representation to model the ge-
ometric evolution of 2D structures along depth. During inference,
the model synthesizes the target slice by directly conditioning on
the relative temporal coordinate corresponding to the desired in-
terpolation position.