While there already exist a number of 2D and 3D pose estimation models with high accuracy, in special domains like sports, which usually require even higher accuracy, there are still spaces to be improved. Existing pose models primarily focus on regular daily activities, which, when being applied to precision sports, such as golf swings, still face limitations. In fact, the rare poses and self-occlusions in golf swing videos can easily mislead regular pose models. To overcome these challenges, we develop a small (2D and 3D) GolfSwing dataset that includes both golfer and club poses. We then fine-tune state-of-the-art 2D and 3D posture models, including HRNet, ViTPose, DEKR, and MixSTE, by GolfSwing into a set of models called GolfPose for golfer-club pose estimation with much higher accuracy. Such a simple-yet-effective method may be generalized to other sports with self-occluded properties. Code is available at https://github.com/MingHanLee/GolfPose .