We present photometric models of 532 disk galaxies in 3.6{\mu}m images from the Spitzer Survey of Stellar Structure in Galaxies (S$^4$G) using the non-parametric DiskFit algorithm. We first test DiskFit's performance on 400 synthetic S$^4$G-like galaxy images. DiskFit is unreliable in the bulge region, but accurately disentangles exponential disks from Ferrers bars farther out as long as their position angles differ by more than 5${^\circ}$. We then proceed to model the S$^4$G galaxies, successfully fitting 489 of them using an automated approach for initializing DiskFit, optimizing the model and deriving uncertainties using a bootstrap-resampling technique. The resulting component geometries and surface brightness profiles are compared to those derived by Salo et al. (2015) using the parametric model galfit. We find generally good agreement between the models, but discrepancies between best-fitting values for individual systems are often significant: the choice of algorithm clearly impacts the inferred disk and bar structure. In particular, we find that DiskFit typically assigns more light to the bar and less light to the disk relative to the Ferrers and exponential profiles presented by Salo et al. (2015) in the bar region. Given DiskFit's reliability at disentangling these components in our synthetic images, we conclude that the surface brightness distributions of barred S$^4$G galaxies are not well-represented by these functional forms. The results presented here underscore the importance of validating photometric decomposition algorithms before applying them to real data and the utility of DiskFit's non-parametric approach at measuring the structure of disks and bars in nearby galaxies.