Our recent work on mitral and tricuspid valve tracking in cardiovascular magnetic resonance (CMR) imaging to obtain accurate evaluations of longitudinal myocardial valve motion (both relaxation and contraction) has enabled an automated diastolic function assessment (e') with CMR. Its time-resolved capability allows a further evaluation of the valve dynamics by providing valve dimension measurements, which are essential to define the etiologies and mechanisms of valve regurgitation. In this paper, we extended the framework to automatically measure mitral annular (MA) and tricuspid annular (TA) dimensions in CMR long-axis cines with a residual neural network backbone. The framework is able to measure MA and TA diameters with an overall excellent accuracy (mean ICC=0.92), on par with an evaluated inter-observer variability (mean ICC=0.92), and to distinguish valvular dimensions between healthy controls and patients with chronic heart failure (p<0.001). Dimension measurements may benefit patients requiring annular sizing and planning of valvular interventions.