Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Image Space Tessellation

Youyang Du, Junqiu Zhu, Zheng Zeng, Lu Wang, Lingqi Yan

TL;DR

This work is the first work to reformulate tessellation as a post-processing operation, shifting it from a pre-rendering geometry pipeline to a screen space neural post-processing stage, and is well-suited for large-scale real-time rendering scenarios.

Abstract

We present Neural Image-Space Tessellation (NIST), a lightweight screen-space post-processing approach that produces the visual effect of tessellated geometry while rendering only the original low-polygon meshes. Inspired by our observation from Phong tessellation, NIST leverages the discrepancy between geometric normals and shading normals as a minimal, view-dependent cue for silhouette refinement. At its core, NIST performs multi-scale neural tessellation by progressively deforming image-space contours with convolutional operators, while jointly reassigning appearance information through an implicit warping mechanism to preserve texture coherence and visual fidelity. Experiments demonstrate that our approach produces smooth, visually coherent silhouettes comparable to geometric tessellation, while incurring a constant per-frame cost and fully decoupled from geometric complexity, making it well-suited for large-scale real-time rendering scenarios. To the best of our knowledge, our NIST is the first work to reformulate tessellation as a post-processing operation, shifting it from a pre-rendering geometry pipeline to a screen space neural post-processing stage.

Neural Image Space Tessellation

TL;DR

This work is the first work to reformulate tessellation as a post-processing operation, shifting it from a pre-rendering geometry pipeline to a screen space neural post-processing stage, and is well-suited for large-scale real-time rendering scenarios.

Abstract

We present Neural Image-Space Tessellation (NIST), a lightweight screen-space post-processing approach that produces the visual effect of tessellated geometry while rendering only the original low-polygon meshes. Inspired by our observation from Phong tessellation, NIST leverages the discrepancy between geometric normals and shading normals as a minimal, view-dependent cue for silhouette refinement. At its core, NIST performs multi-scale neural tessellation by progressively deforming image-space contours with convolutional operators, while jointly reassigning appearance information through an implicit warping mechanism to preserve texture coherence and visual fidelity. Experiments demonstrate that our approach produces smooth, visually coherent silhouettes comparable to geometric tessellation, while incurring a constant per-frame cost and fully decoupled from geometric complexity, making it well-suited for large-scale real-time rendering scenarios. To the best of our knowledge, our NIST is the first work to reformulate tessellation as a post-processing operation, shifting it from a pre-rendering geometry pipeline to a screen space neural post-processing stage.
Paper Structure (12 sections, 4 figures, 2 tables)

This paper contains 12 sections, 4 figures, 2 tables.

Table of Contents

  1. RESULTS AND COMPARISONS
  2. Qualitative and Quantitative Comparison
  3. Resolution Scaling and Performance
  4. Ablation Study
  5. Discussion and Limitations
  6. Generality.
  7. Screen Space Limitations.
  8. Performance.
  9. Quality.
  10. Shading Normals.
  11. Geometric Normals vs. Depth.
  12. CONCLUSION AND FUTURE WORK

Figures (4)

  • Figure 5: Screen space artifacts caused by partially visible triangles. In this case, incomplete geometric information may lead to unstable image space deformation and visible artifacts.
  • Figure 6: Qualitative comparison across four scenes. From top to bottom: SoulCave, Cowboy, Bronze, and Junkyard. The left column shows the output of NIST. For each scene, two representative regions are highlighted and magnified on the right, comparing the input rendering, our result, and the reference. We also show the geometric normals and shading normals in the middle. NIST effectively removes silhouette artifacts while producing results visually comparable to geometric tessellation.
  • Figure 7: Ablation study of NIST. From left to right: input rendering, variants without implicit deformation, without feature warping, without LPIPS loss, our full model, and the PN-Triangles reference. Removing the implicit deformation module prevents effective silhouette refinement. Without feature warping, visible seams and appearance discontinuities emerge in newly deformed regions (highlighted by red arrows). Excluding LPIPS leads to overly smooth and blurred results. The bottom row shows FLIP error maps with respect to the reference, where lower values indicate better perceptual similarity.
  • Figure 8: Preservation of straight element guided by normal consistency. In regions where geometric normals and shading normals agree (middle), NIST correctly avoids deformation and preserves straight structures (left, zoom-in), while PN-Triangles introduces undesired deformation (right, zoom-in).