Individualised head-related transfer functions (HRTFs) are essential for creating realistic virtual reality (VR) and augmented reality (AR) environments and interactions. Performing acoustic measurements is the most accurate way to capture these individualised HRTFs. However, one of the main challenges is acoustically capturing high-quality HRTFs without the need for expensive equipment and a lab- controlled setting. To make these measurements more feasible on a large scale, HRTF upsampling has been exploited in the past, where a high-resolution HRTF is created from a low-resolution measurement. However, as the world shifts to more data-driven methods, upsampling HRTFs using machine learning (ML) has become more prevalent. The main limitation is the lack of HRTF data available for model training. This paper explores the use of transfer learning (TL) on a new synthetic HRTF dataset generated from three-dimensional (3D) meshes using a parametric pina model, and the performance improvements that can be achieved. The performance is also compared against using a small acoustically measured dataset for TL, with the aim to start answering the question: ‘Is it better to have more data, or is it better for the data to be of higher quality?’.