zju3dv/manhattan_sdf
A neural network-based 3D scene reconstruction method that leverages Manhattan-world structural assumptions for improved geometric accuracy.
Not currently ranked — collecting fresh signals.
star history
This repository implements a CVPR 2022 Oral paper on neural 3D reconstruction using signed distance functions (SDF) under the Manhattan-world assumption. The method reconstructs 3D scenes from multi-view RGB images by combining implicit neural representations with geometric priors about structural regularity in man-made environments. It provides training and evaluation code for the ScanNet dataset.
Frequently asked
- What is zju3dv/manhattan_sdf?
- A neural network-based 3D scene reconstruction method that leverages Manhattan-world structural assumptions for improved geometric accuracy.
- Is manhattan_sdf open source?
- Yes — zju3dv/manhattan_sdf is an open-source project tracked on heatdrop.
- What language is manhattan_sdf written in?
- zju3dv/manhattan_sdf is primarily written in Python.
- How popular is manhattan_sdf?
- zju3dv/manhattan_sdf has 532 stars on GitHub.
- Where can I find manhattan_sdf?
- zju3dv/manhattan_sdf is on GitHub at https://github.com/zju3dv/manhattan_sdf.