Hexels
A novel-view synthesis and 3D reconstruction method based on planar hexagonal primitives, each carrying a first-order spatial color field, achieving efficient scene representation with fewer primitives.
Timeline (last 90 days)
Hexels replaces traditional circular or Gaussian primitives with planar hexagonal primitives, each carrying a first-order spatial color field.
In tests across 34 scenes, Hexels improves PSNR by 4.53–7.38 dB over memory-matched 2DGS while using lower training memory.
On the 3D surface reconstruction task, Hexels reduces error by 28.9% compared with memory-matched 2DGS across 15 DTU scans.
On 6 Tanks and Temples surface scenes, Hexels achieves the highest average F1 score (34.38%) among the compared methods, versus 31.72% for 2DGS.
Hexels outperforms Image-GS on the 2D image representation task.
Hexels belongs to the surface-aligned primitives technical line and, like methods such as 2DGS, falls in the planarized primitive direction.
Hexels emphasizes a complementary quality-memory trade-off and does not fully replace existing methods in all scenarios.
The local planarity assumption of hexagonal primitives may face challenges in regions with dense curvature or high-frequency detail, which requires follow-up work to verify.
The current material does not disclose specific figures for Hexels' training time or inference memory, nor does it describe performance in highly non-planar or texture-complex scenes.
All Facts (9)
The current material does not disclose specific figures for Hexels' training time or inference memory, nor does it describe performance in highly non-planar or texture-complex scenes.
50%UnverifiedIn tests across 34 scenes, Hexels improves PSNR by 4.53–7.38 dB over memory-matched 2DGS while using lower training memory.
50%UnverifiedOn the 3D surface reconstruction task, Hexels reduces error by 28.9% compared with memory-matched 2DGS across 15 DTU scans.
50%UnverifiedOn 6 Tanks and Temples surface scenes, Hexels achieves the highest average F1 score (34.38%) among the compared methods, versus 31.72% for 2DGS.
50%UnverifiedHexels outperforms Image-GS on the 2D image representation task.
50%UnverifiedHexels belongs to the surface-aligned primitives technical line and, like methods such as 2DGS, falls in the planarized primitive direction.
50%UnverifiedHexels emphasizes a complementary quality-memory trade-off and does not fully replace existing methods in all scenarios.
50%UnverifiedHexels replaces traditional circular or Gaussian primitives with planar hexagonal primitives, each carrying a first-order spatial color field.
50%UnverifiedThe local planarity assumption of hexagonal primitives may face challenges in regions with dense curvature or high-frequency detail, which requires follow-up work to verify.
50%