Abstract: Complex oxide thin films exhibit unique and useful properties for electronics, energy, communications, and more. Imaging the atomic-scale structure of these films is crucial for deducing and ultimately engineering their functional behavior, but standard x-ray diffraction techniques suffer from the phase retrieval problem, which is exacerbated for nanometer sized films. Current approaches analyze crystal truncation rod (CTR) diffraction using constrained iterative algorithms to output a 3D electron density to obtain the structure. Unfortunately, state-of-the-art methodologies are typically heavily dependent on initial guesses, require high data density, and fail for thick films. Here, we propose and demonstrate a new machine learning-based phase retrieval technique for thin films – Machine Learning for Material Bragg-rod Analysis (MAMBA). MAMBA is based on a U-Net architecture that takes in the measured CTR intensity as input, and outputs the complex scattered electric field, from which the electron density $\rho(\vec r)$ can be obtained by Fourier inversion. We summarize the promising results from MAMBA using simulated data, showing its potential for providing high-precision atomic structures of thin films beyond limitations of standard phase-retrieval techniques.