<p>Estimating the 6D pose of objects from a single RGB image is important for various applications such as robot vision, autonomous driving, and AR/VR. One common approach is to first estimate the initial pose and then refine it based on the optical flow between a target image and a reference image rendered from the initial pose. However, current methods face challenges in accurately estimating flow in low-texture regions and under large displacements. To address these issues, we propose a new approach that uses the geometric information of the target object to improve optical flow and enhance pose estimation performance. Our method introduces a probabilistic flow estimation model to predict a confidence map, which guides the newly proposed Adaptive Flow-Geo Fusion (AFGF) module. The AFGF module intelligently combines flow and geometric features, leading to more accurate flow refinement and, consequently, more precise pose estimation. Our method achieves a +5.4% gain in ADD(-S) accuracy on YCB-V and a +1.0% gain in mean mAR on YCB-V and LM-O compared to the previous state-of-the-art. Comprehensive experiments on benchmark datasets demonstrate the superiority of our approach.</p>