.png){kind=icon}
.png){kind=icon}
Return
Hash-NURF: efficient nested transparent object reconstruction using multi-resolution hash encoding
.png){kind=icon}
PRE
AI
DOI：10.1007/s00371-026-04394-1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Transparent object reconstruction is crucial in various domains, including 3D modeling and robotic perception. However, accurately reconstructing nested transparent objects remains challenging due to complex light interactions. Recent advances like NU-NeRF have shown promise but suffer from computational inefficiency. Here, we introduce Hash-NURF, a framework that integrates multi-resolution hash encoding with neural implicit representations to accelerate reconstruction. Specifically, we utilize hash-based NeRF and SDF for efficient optimization. To unbounded scenes, we added a contractive mapping strategy. We also incorporated special sampling and regularization techniques to improve the model’s stability and help it converge faster. Our method significantly reduces training time by up to three times while maintaining comparable reconstruction quality. We validate our approach on synthetic datasets, demonstrating its efficacy in both outer and inner surface reconstruction of nested transparent objects. Visit our repository at https://github.com/SyouSanGin/Hash-NURF .
Keywords:
Signed distance field (SDF)
Transparent object
Reconstruction
Neural implicit representations
Journal
T
IF:
0
Papers:
369
Citations:
0