The field of 3D scene reconstruction and localization is witnessing significant advancements, driven by the development of novel methods that leverage semantic information and Gaussian splatting techniques. Researchers are exploring new approaches to reconstruct scenes from unposed images, incorporating semantic rays and features to improve localization accuracy. A key direction in this field is the integration of 3D Gaussian splatting with registration techniques, enabling the alignment of local Gaussians and the estimation of camera poses. Another important trend is the use of semantic information to guide the segmentation and localization process, leading to improved performance and generalization ability. Notably, the development of joint registration modules and multi-level pose regression strategies is allowing for more accurate and efficient localization. Overall, these innovations are pushing the boundaries of 3D scene reconstruction and localization, with potential applications in various fields such as computer vision and robotics. Noteworthy papers include: RegGS, which proposes a 3D Gaussian registration-based framework for reconstructing unposed sparse views. Supercharging Floorplan Localization with Semantic Rays, which introduces a semantic-aware localization framework that jointly estimates depth and semantic rays. SGLoc, which proposes a novel localization system that directly regresses camera poses from 3D Gaussian Splatting representation by leveraging semantic information.