Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:BGVVDENOrecord.jsonopen to challenge →
read the original abstract
With the growing demand for high-fidelity 3D models from 2D images, existing methods still face significant challenges in accurately reproducing fine-grained geometric details due to limitations in domain gaps and inherent ambiguities in RGB images. To address these issues, we propose Hi3DGen, a novel framework for generating high-fidelity 3D geometry from images via normal bridging. Hi3DGen consists of three key components: (1) an image-to-normal estimator that decouples the low-high frequency image pattern with noise injection and dual-stream training to achieve generalizable, stable, and sharp estimation; (2) a normal-to-geometry learning approach that uses normal-regularized latent diffusion learning to enhance 3D geometry generation fidelity; and (3) a 3D data synthesis pipeline that constructs a high-quality dataset to support training. Extensive experiments demonstrate the effectiveness and superiority of our framework in generating rich geometric details, outperforming state-of-the-art methods in terms of fidelity. Our work provides a new direction for high-fidelity 3D geometry generation from images by leveraging normal maps as an intermediate representation.
This paper has not been read by Pith yet.
Forward citations
Cited by 18 Pith papers
-
Sculpting NeRF Geometry: Human-Preference Fine-Tuning of a 3D-Aware Face GAN
Fine-tunes EG3D using a human-preference reward on NeRF density to improve face geometry, achieving 74.4% user preference in pairwise tests with FID rising from 4.09 to 6.66.
-
VDE: Training-Free Accelerating Rectified Flow Model via Velocity Decomposition and Estimation
VDE accelerates rectified flow models like Flux by 3.22x with LPIPS of 0.069 via velocity decomposition into parallel/orthogonal components plus periodic full-pass anchoring.
-
3DReflecNet: A Large-Scale Dataset for 3D Reconstruction of Reflective, Transparent, and Low-Texture Objects
3DReflecNet is a 22 TB+ dataset of over 120,000 synthetic and 1,000 real objects with millions of multi-view frames for benchmarking 3D reconstruction on reflective, transparent, and low-texture surfaces.
-
Mix3R: Mixing Feed-forward Reconstruction and Generative 3D Priors for Joint Multi-view Aligned 3D Reconstruction and Pose Estimation
Mix3R mixes feed-forward reconstruction and generative 3D priors via Mixture-of-Transformers and overlap-based attention bias to achieve better-aligned 3D shapes and more accurate poses than either approach alone.
-
Make-It-Poseable: Feed-forward Latent Posing Model for 3D Characters
A latent-space transformer framework poses 3D characters without skinning or fixed topologies, outperforming baselines and generalizing zero-shot to quadrupeds.
-
ELSA3D: Elastic Semantic Anchoring for Unified 3D Understanding and Generation
ELSA3D introduces elastic semantic anchoring via sparse anchor tokens and a scale-aware octree tokenizer to unify 3D generation and captioning at reduced computational cost.
-
ROAR-3D: Routing Arbitrary Views for High-Fidelity 3D Generation
ROAR-3D adds a token-wise view router and dual-stream attention to pretrained single-view 3D generators so they can use arbitrary unposed images for higher-fidelity output.
-
Pixal3D: Pixel-Aligned 3D Generation from Images
Pixal3D performs pixel-aligned 3D generation from images via back-projected multi-scale feature volumes, achieving fidelity close to reconstruction while supporting multi-view and scene synthesis.
-
A Few-Step Generative Model on Cumulative Flow Maps
Cumulative flow maps unify few-step generative modeling for diffusion and flow models via cumulative transport and parameterization with minimal changes to time embeddings and objectives.
-
Grasp in Gaussians: Fast Monocular Reconstruction of Dynamic Hand-Object Interactions
GraG reconstructs dynamic 3D hand-object interactions from monocular video 6.4x faster than prior work by using compact Sum-of-Gaussians tracking initialized from large models and refined with 2D losses.
-
ReplicateAnyScene: Zero-Shot Video-to-3D Composition via Textual-Visual-Spatial Alignment
ReplicateAnyScene performs fully automated zero-shot video-to-compositional-3D reconstruction by cascading alignments of generic priors from vision foundation models across textual, visual, and spatial dimensions.
-
UniRecGen: Unifying Multi-View 3D Reconstruction and Generation
UniRecGen unifies reconstruction and generation via shared canonical space and disentangled cooperative learning to produce complete, consistent 3D models from sparse views.
-
Native and Compact Structured Latents for 3D Generation
Introduces O-Voxel omni-voxel representation and Sparse Compression VAE for structured native 3D latents, enabling efficient training of large flow-matching models that produce higher-quality geometry and materials th...
-
InfiniVerse: Occupancy Guided Unbounded Scene Generation for Autonomous Driving
InfiniVerse reconstructs 3D occupancy from one frame, extends scenes autoregressively, converts to video via diffusion, and uses re-projection feedback to achieve SOTA FID 6.4 and FVD 67.97 on Waymo and nuScenes.
-
ReScene: Structured Indoor Scene Reconstruction from Multi-View Captures
ReScene introduces HierView for view prioritization and Relation-Aware Assembly for scene graph fusion, reporting 17% lower Chamfer Distance and 26% lower LPIPS than prior baselines on ScanNet while running faster.
-
MM-TRELLIS: Point-Cloud Guided Multi-Modal 3D Vehicle Generation in Autonomous Driving
MM-TRELLIS extends TRELLIS with LiDAR point-cloud guidance and multi-view image conditioning plus voxel filtering to generate high-fidelity 3D vehicle meshes from in-the-wild driving data.
-
WorldAct: Activating Monolithic 3D Worlds into Interactive-Ready Object-Centric Scenes
WorldAct activates monolithic 3D worlds into interactive scenes via multimodal agent-guided decomposition, geometrically aligned mesh reconstruction, and 3D inpainting.
-
AnimateAnyMesh++: A Flexible 4D Foundation Model for High-Fidelity Text-Driven Mesh Animation
AnimateAnyMesh++ animates arbitrary 3D meshes from text using an expanded 300K-identity DyMesh-XL dataset, a power-law topology-aware DyMeshVAE-Flex, and a variable-length rectified-flow generator to produce semantica...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.