The field of 3D generation and reconstruction is rapidly advancing, with a focus on developing innovative methods for creating accurate and detailed 3D models from various inputs. Recent research has explored the use of diffusion models, geometric constraints, and multi-modal guidance to improve the quality and controllability of 3D generation. Notable developments include the use of epipolar geometry to enhance video generation models, and the introduction of novel frameworks for generating high-fidelity 3D meshes, articulated objects, and 3D human poses from sketches. These advancements have significant implications for applications in computer-aided design, robotics, and animation.

Noteworthy papers include: Sketch2BIM, which introduces a human-in-the-loop pipeline for converting hand-drawn floor plans to 3D BIM models. Topology Sculptor, Shape Refiner, which presents a novel method for generating high-quality 3D meshes using discrete diffusion models. ArtiLatent, which proposes a generative framework for synthesizing realistic articulated 3D objects with fine-grained geometry and accurate articulation.