Generating 3D Models for Prototyping of Virtual Environments using NeRF
📜 Abstract
This research focuses on optimizing 3D modeling by leveraging class-specific structures to enhance the speed and accuracy of 3D reconstruction. The proposed model integrates SlowFast CNN for structural feature extraction and Edge Collapsing for outlier rejection and complexity reduction. It generates 3D point clouds, identifies consistent structures, and constructs refined meshes. Key technologies used include Neural Radiance Fields (NeRF), Structure From Motion (SfM), SlowFast CNN, Edge Collapsing, and CLIP for evaluation, achieving a CLIP Similarity Score of 0.542 with an average execution time of 3.27 minutes.
✨ Summary
Summary
The paper proposes a NeRF-based pipeline for generating 3D models intended for virtual-environment prototyping. Its workflow combines Structure-from-Motion for reconstruction support, SlowFast CNN for extracting structural features, Edge Collapsing for removing outliers and reducing mesh complexity, and CLIP similarity for semantic evaluation. The reported evaluation achieved a CLIP similarity score of 0.542 with an average execution time of 3.27 minutes. (in.linkedin.com)
The available citation evidence indicates limited but concrete subsequent uptake. The work is cited in later research on reconstructing real-world environments for virtual driving and in research on LiDAR-based reproduction of road scenarios for simulation. These citations treat the paper primarily as an example of NeRF-based 3D-environment generation rather than as a central methodological foundation. (ouci.dntb.gov.ua)
No documented industrial deployment or commercial product adoption directly attributable to this paper was identified in the available search results.