We propose GTrans, a transfer learning framework that enhances graphon estimation for small graphs via neighborhood smoothing, Gromov-Wasserstein transport, and adaptive debiasing, boosting link prediction and graph classification.