This paper presents two novel likelihood terms for silhouettes and contours in model-based contexts. Despite the power of such formulations, building likelihoods that truly reflect the good configurations of the problem is by no means easy due to, most commonly,