Search papers, labs, and topics across Lattice.
This paper adapts 3D Gaussian Splatting (3DGS) to improve 3D reconstruction from sparse-view Annular Dark Field Scanning Transmission Electron Microscopy (ADF-STEM) tomography. They introduce a learnable scalar field, denza, to model local scattering strength, a coefficient 纬 for scattering view normalization, and a 2D Fourier amplitude loss to suppress missing wedge artifacts. Experiments on 45-view and 15-view tilt series demonstrate that their "DenZa-Gaussian" method achieves higher fidelity reconstructions and more accurate 2D projections compared to existing methods, particularly under sparse-view conditions.
Achieve high-fidelity 3D reconstructions from sparse-view electron microscopy data by adapting 3D Gaussian Splatting, enabling accurate analysis of dose-sensitive materials.
Analytical Dark Field Scanning Transmission Electron Microscopy (ADF-STEM) tomography reconstructs nanoscale materials in 3D by integrating multi-view tilt-series images, enabling precise analysis of their structural and compositional features. Although integrating more tilt views improves 3D reconstruction, it requires extended electron exposure that risks damaging dose-sensitive materials and introduces drift and misalignment, making it difficult to balance reconstruction fidelity with sample preservation. In practice, sparse-view acquisition is frequently required, yet conventional ADF-STEM methods degrade under limited views, exhibiting artifacts and reduced structural fidelity. To resolve these issues, in this paper, we adapt 3D GS to this domain with three key components. We first model the local scattering strength as a learnable scalar field, denza, to address the mismatch between 3DGS and ADF-STEM imaging physics. Then we introduce a coefficient $纬$ to stabilize scattering across tilt angles, ensuring consistent denza via scattering view normalization. Finally, We incorporate a loss function that includes a 2D Fourier amplitude term to suppress missing wedge artifacts in sparse-view reconstruction. Experiments on 45-view and 15-view tilt series show that DenZa-Gaussian produces high-fidelity reconstructions and 2D projections that align more closely with original tilts, demonstrating superior robustness under sparse-view conditions.
