The field of 3D generation and scene understanding is rapidly evolving, with a focus on developing more efficient and effective methods for generating high-quality 3D models and understanding complex scenes. Recent research has explored the use of large language models, generative adversarial networks, and diffusion models to improve the accuracy and fidelity of 3D generation. Additionally, there is a growing interest in developing methods that can understand and generate 3D scenes in a more human-like way, such as by incorporating knowledge of object relationships and scene semantics. Noteworthy papers in this area include DreamLLM-3D, which presents a novel approach for affective dream reliving using large language models and 3D generative AI, and HSM, which introduces a hierarchical framework for indoor scene generation with dense object arrangements. Other notable papers include RelTriple, which enhances furniture distribution by learning spacing relationships between objects and regions, and SparseFlex, which enables differentiable mesh reconstruction at high resolutions directly from rendering losses.
Advances in 3D Generation and Scene Understanding
Sources
DreamLLM-3D: Affective Dream Reliving using Large Language Model and 3D Generative AI
iFlame: Interleaving Full and Linear Attention for Efficient Mesh Generation
HSM: Hierarchical Scene Motifs for Multi-Scale Indoor Scene Generation
PVChat: Personalized Video Chat with One-Shot Learning
Geometric Constrained Non-Line-of-Sight Imaging
Decorum: A Language-Based Approach For Style-Conditioned Synthesis of Indoor 3D Scenes
Global-Local Tree Search for Language Guided 3D Scene Generation
Training-Free Personalization via Retrieval and Reasoning on Fingerprints
MC-LLaVA: Multi-Concept Personalized Vision-Language Model
Learning 3D Object Spatial Relationships from Pre-trained 2D Diffusion Models
RelTriple: Learning Plausible Indoor Layouts by Integrating Relationship Triples into the Diffusion Process
MAR-3D: Progressive Masked Auto-regressor for High-Resolution 3D Generation
SparseFlex: High-Resolution and Arbitrary-Topology 3D Shape Modeling
CTRL-O: Language-Controllable Object-Centric Visual Representation Learning
Shape Generation via Weight Space Learning
Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging
Mono2Stereo: A Benchmark and Empirical Study for Stereo Conversion