The field of 3D reconstruction and avatar modeling is moving towards more efficient and accurate methods for handling large-scale scenes and generating high-quality 3D models from limited input data. Recent developments have focused on improving the scalability and robustness of scene regression methods, as well as enhancing the fidelity of 3D avatar reconstruction. Notable advancements include the use of feed-forward transformer architectures and novel encoder-decoder neural network designs to achieve fast and robust model inference. Additionally, there is a growing interest in addressing the challenges of reconstructing complete and consistent 3D avatars, particularly in regards to modeling the back-head regions. Overall, the field is witnessing significant progress towards more realistic and interactive 3D modeling applications. Noteworthy papers include: SAIL-Recon, which introduces a feed-forward transformer for large-scale structure-from-motion, and FastAvatar, which proposes a unified framework for fast high-fidelity 3D avatar reconstruction. AvatarBack is also notable for its novel approach to reconstructing complete 3D avatars by explicitly modeling the missing back-head regions.