This paper presents our experiments to quantify the manifolds learned by machine learning models, specifically using a Generative Adversarial Network (GAN). We compare the manifolds learned at each epoch to the real manifolds representing the actual data. To quantify a manifold, we study the intrinsic dimensions and topological features of the manifold learned by the ML model, observing how these metrics evolve during training and whether they converge to the metrics of the real data manifold. We address the challenges of evaluating GANs beyond traditional loss metrics. Commonly, the quality of GAN outputs is judged by human perception, which can be subjective and overlook issues like mode collapse. Our research proposes an alternative evaluation method that leverages intrinsic dimension and topological features of the data manifold. By comparing these features between generated and real datasets, we can objectively assess GAN performance and determine optimal stopping points for training. Our findings indicate that these topological metrics align with visual inspection and the Fréchet Inception Distance (FID) score, suggesting a more efficient training process and resource utilization.