The precision of cardiac magnetic resonance segmentation is an important area to investigate clinically and has received a lot of attention from the research community for its impact on the evaluation of cardiac functions. However, the correct identification of key time frames of cardiac sequences has received significantly less attention, especially in the MR domain, despite its great importance in the correct measurement of the Ejection Fraction, a key metric in diagnostics. In this paper, we present two deep learning regression methods to automate the otherwise time-consuming annotation process, with performance within the 1–2 frame distance error and almost instant calculation over short-axis images from a public dataset. Results are presented using publicly available data.