Mode group diversity multiplexing (MGDM)-based multimode fiber (MMF) systems have been known to be promising solutions for high-data-rate links. However, the challenge of cross-talk within these systems remains a critical concern, since coupling across mode groups diminishes data rates. Past work has shown that optimizing launch conditions and effective spatial filtering can minimize cross-talk across mode groups, although the optimization of the launching has been largely using trial-and-error-based approaches. Characterizing cross-talk limits and quantifying mode coupling requires the use of coherent imaging of the fiber, which is prohibitively complex. In this paper, we present, to our knowledge, a novel approach to characterize the received spatial signal at the MMF output, by obtaining both the amplitude and phase, using just imaging intensity measurements. This is enabled by leveraging recent advances in signal processing to recover the phase from intensity measurements using an alternating minimization-based algorithm for low-complexity phase retrieval, and presents an accurate characterization of mode coupling and estimation of cross-talk in MMF MGDM systems. Experimental validation demonstrates the effectiveness of this approach, particularly in offset-launch-based large-core MMF MGDM systems, where optimal launch parameters are identified to minimize cross-talk. Our findings highlight the significance of optimal launching and spatial filtering in maximizing mode separation, thus reducing cross-talk and enhancing overall system performance, as validated through extensive data rate experiments. The novel phase-retrieval-based fiber characterization can be extended and used to efficiently design spatial filtering solutions for MGDM, such as photonic lanterns and mode multiplexers.