DualPrim: Compact 3D Reconstruction with Positive and Negative Primitives
Authors: Xiaoxu Meng, Zhongmin Chen, Bo Yang, Weikai Chen, Weixiao Liu, Lin Gao
Organizations: Independent Researcher · Institute of Computing Technology, Chinese Academy of Sciences · School of Advanced Interdisciplinary Science, University of Chinese Academy of Sciences · Waymo LLC · Lucid Motors
Neural reconstructions often trade structure for fidelity, yielding dense and unstructured meshes with irregular topology and weak part boundaries that hinder editing, animation, and downstream asset reuse. We present DualPrim, a compact and structured 3D reconstruction framework. Unlike additive-only implicit or primitive methods, DualPrim represents shapes with positive and negative superquadrics: the former builds the bases while the latter carves local volumes through a differentiable operator, enabling topology-aware modeling of holes and concavities. This additive-subtractive design increases the representational power without sacrificing compactness or differentiability. We embed DualPrim in a volumetric differentiable renderer, enabling end-to-end learning from multi-view images and seamless mesh export via closed-form boolean difference. Empirically, DualPrim delivers state-of-the-art accuracy and produces compact, structured, and interpretable outputs that better satisfy downstream needs than additive-only alternatives.
Figures & tables
Figure 2: The first row shows the artist-created meshes (AMs), and the second row shows the exportation from Marching Cubes [ 30 ] .
Figure 3: The overview of DualPrim pipeline. We represent the 3D scene as a set of primitive parameters and optimize them within a differentiable rendering framework under multi-view image supervision. The final mesh is obtained by computing the Boolean difference between the positive and negative superquadric components of each primitive.
Parameter Name
Shape
Explanation
Range
scale parameters (ax,ay,az)
(2,3)
The radius of the x-, y-, z-axis of PSQ and NSQ.
[0.02,1.0]
shape parameters (ϵ1,ϵ2)
(2,2)
The parameters controlling the shape of PSQ and NSQ.
[0.05,2]
primitive transparency α
(1)
The transparency of the super-primitive.
[0,1]
primitive render sharpness θ
(1)
The sharpness of the super-primitive.
[0,1]
translation T
(2,3)
The translation of the PSQ and the NSQ.
[−1,1]
rotation R
(2,3)
The rotation angles of the PSQ and the NSQ.
[−180,180)
Table 1: The full parameter list of a super primitive.
Figure 4: Illustration of the SDF calculation with PSQ and NSQ.
Figure 5: We show the rendering of three example dual-primitives.
Figure 6: Qualitative comparison on in-the-wild objects. Each object occupies two rows. The first row shows input, Ours, PA, MA, MP, EMS, and NVD. The second row shows FC, RNS, CAPRI, D 2 CSG, DiffCSG, 2DGS+MC, and 2DGS+CSG.
Figure 7: The ablation study on regularization terms.
Method
#V(k) ↓
#F(k) ↓
NC Loss ↓
CD ↓
EditScore ↑
RNS [ 7 ]
124.41
57.44
37.45
10.71
4.77
NVD [ 36 ]
15.40
13.98
56.19
10.63
5.47
FC [ 47 ]
16.67
8.33
16.11
13.32
4.98
PA [ 62 ]
21.51
12.61
4.48
11.57
4.59
MP [ 28 ]
3.38
1.75
12.42
12.58
4.88
MA [ 10 ]
0.55
0.29
4.00
10.12
5.12
Table 2: Quantitative comparison and user evaluation of DualPrim (ours) with 13 representative methods.
Meshes remain the most practical representation for geometry reasoning and integration into graphics pipelines, yet existing reconstruction methods struggle to produce high-quality meshes. Most state-of-the-art approaches initially learn an intermediate representation (NeRF/3DGS) and treat mesh extraction as a post-processing step, which often leads to oversmoothed surfaces or poor quality meshes with excessive triangle counts.Existing mesh-native optimization methods alleviate some of these issues but suffer from fixed-resolution discretizations and unstable optimization behavior. In this paper, we introduce an adaptive mesh-based optimization framework and a practical mesh rendering technique to address these challenges. Our representation combines an optimizable Delaunay-triangulated tetrahedral grid with a multi-resolution hash grid. The former is refined through point pruning and insertion, while the latter provides latent features for SDF/appearance value predictions. We use volumetric rendering to bootstrap a coarse geometry while leveraging mesh-based rendering for recovering fine-grained details. Additionally, we propose a differentiable, rasterization-based depth-offset rendering formulation, reducing geometric artifacts and improving reconstruction quality. Our method significantly outperforms existing mesh optimization approaches across a variety of object-centric benchmarks while being competitive with state-of-the-art NeRF/3DGS methods.
Johannes Weidenfeller, Shaofei Wang, Philipp Fürnstahl +1
ETH Zurich, Zurich, Switzerland · Beijing Institute for General Artificial Intelligence, Beijing, China · University of Zurich, Zurich, Switzerland +1
We present UniQueR, a unified query-based feedforward framework for efficient and accurate 3D reconstruction from unposed images. Existing feedforward models such as DUSt3R, VGGT, and AnySplat typically predict per-pixel point maps or pixel-aligned Gaussians, which remain fundamentally 2.5D and limited to visible surfaces. In contrast, UniQueR formulates reconstruction as a sparse 3D query inference problem. Our model learns a compact set of 3D anchor points that act as explicit geometric queries, enabling the network to infer scene structure, including geometry in occluded regions--in a single forward pass. Each query encodes spatial and appearance priors directly in global 3D space (instead of per-frame camera space) and spawns a set of 3D Gaussians for differentiable rendering. By leveraging unified query interactions across multi-view features and a decoupled cross-attention design, UniQueR achieves strong geometric expressiveness while substantially reducing memory and computational cost. Experiments on Mip-NeRF 360 and VR-NeRF demonstrate that UniQueR surpasses state-of-the-art feedforward methods in both rendering quality and geometric accuracy, using an order of magnitude fewer primitives than dense alternatives.
Chensheng Peng, Quentin Herau, Jiezhi Yang +6
Applied Intuition · UC Berkeley · Stanford University
We present SHELLS (Semantic Head Estimation via Layered Local Sampling), an efficient feed-forward framework for 3D head reconstruction in dense semantic correspondence from multi-view images. Existing methods typically refine vertices independently via localized feature volumes. This approach couples memory-intensive feature sampling to mesh resolution, which limits scalability for dense topologies (> 10k vertices) and introduces surface noise. In contrast, SHELLS decouples feature extraction from mesh resolution via a hierarchical sampling strategy. We extract multi-view features using a DINOv2 backbone with LoRA adaptation, projectively sample a sparse global feature cloud, and predict an intermediate coarse mesh. This coarse prior guides the construction of layered, surface-aware sampling shells that serve as a discrete search space for the final reconstruction. SHELLS maintains surface consistency while using 88% less inference GPU memory (2.4GB vs. 20GB) than volumetric baselines. It reduces median registration error by 21% to 29% with a 3.5x inference speedup (0.08s vs. 0.29s) for 18k-vertex meshes. Notably, our model is trained exclusively on synthetic data yet generalizes effectively to real-world captures, eliminating the need for the costly, pre-registered multi-view datasets common in prior work.