Force myography has recently gained increasing attention for hand gesture recognition tasks. In this paper, we attempt to improve gesture recognition accuracy by utilising data from multiple persons using transfer learning. We experiment with both a progressive neural network architecture and a variation, a combined progressive neural network which seeks to learn more generalised features from the source domains. We show that while transfer learning can improve performance over a fully-connected neural network, care needs to be taken lest one end up with negative transfer. In the process, we note a lack of publicly available benchmark data in this field, with most existing studies collecting their own data often with custom hardware and for varying sets of gestures. This limits the effectiveness of such transfer learning approaches as well as the ability to compare various algorithms. We therefore also contribute to the advancement of this field by making accessible a benchmark dataset, collected using a commercially available sensor setup from 20 persons covering 18 unique gestures. We hope that others may utilise our data for similar transfer learning approaches, while also allowing further comparison of results and easier entry into the field of force myography gesture recognition.