From 2D to 3D: Leveraging Sparse Inputs for High-Fidelity Model Generation with Neural Radiance Fields

Reeta Koshy; Sakshi Bisen; Arjun Shinde; Hrishabh Upadhyay1

1

Publication Date: 2024/12/05

Abstract: Rendering 2D images into 3D models is a significant challenge in computer vision, with applications ranging from robotics to augmented reality. This paper presents a novel framework leveraging Neural Radiance Fields (NeRF) and its advancements to achieve efficient and high-fidelity 3D reconstruction. Our approach integrates feature extraction, ray sampling, and pose estimation using entropy-based optimization and attention-based aggregation, ensuring robust performance across diverse datasets. Key techniques include using PixelNeRF for few-shot rendering, iNeRF for pose refinement, and General Radiance Fields (GRF) for unseen geometries. Experiments demonstrate superior results in 3D representation accuracy, novel view synthesis, and generalization capabilities. This research highlights the potential of NeRF-based systems to revolutionize 3D modeling and content generation while addressing the limitations of traditional methods.

Keywords: NeRF, 2D-to-3D Rendering, iNeRF, PixelNeRF, General Radiance Fields, Pose Estimation, Few-Shot Learning.

DOI: https://doi.org/10.5281/zenodo.14281148

PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT24NOV1342.pdf

REFERENCES

  1. Mildenhall, B., et al. (2020). NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis. In Proceedings of the European Conference on Computer Vision (ECCV), 2020.
  2. Zhang, Y., et al. (2021). iNeRF: Inverting Neural Radiance Fields for Pose Estimation. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021.
  3. Tancik, M., et al. (2021). GRF: Learning a General Radiance Field for 3D Representation and Rendering.
  4. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.
  5. Yu, Z., et al. (2021). PixelNeRF: Generating 3D Neural Radiance Fields from a Single Image. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021.
  6. Srinivasan, P. P., et al. (2021). NeRF-W: Neural Radiance Fields Without Knowing Camera Poses. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.
  7. Mildenhall, B., et al. (2022). Neural Implicit Representations for 3D Reconstruction. IEEE Transactions on Visualization and Computer Graphics, 28(7), 2001-2014.
  8. Li, S., & Zhang, L. (2023). Efficient 3D Scene Reconstruction from Sparse 2D Views using Neural Radiance Fields. International Journal of Computer Vision, 45(5), 670-680.