HashPoint: Accelerated Point Searching and Sampling for Neural Rendering

CVPR 2024 (Highlight)

Paper | Video | Project Page

The project provides official implementation of HashPoint in CVPR'24. The HashPoint method significantly improves the efficiency of point searching and sampling by combining rasterization with ray-tracing techniques. Unlike the multi-surface sampling approach of Point-NeRF, HashPoint concentrates on the primary surface, thereby acclerating the rendering process. The video above demonstrates this streamlined approach in action.

TODOs

• The source code of hashpoint (Searching and Sampling)
• Integration with Point-NeRF (Coming soon)
• The detailed instructions (Coming soon)

Installation

1. Create a new conda environment named mvchm to run this project:

conda create -n hashpoint python=3.9.7
conda activate hashpoint

2. Make sure your system meet the CUDA requirements and install some core packages:

pip install easydict torch==1.12.1+cu113 torchvision==0.13.1+cu113 tqdm scipy opencv-python

3. Clone this repository

cd Your-Project-Folder
gir clone [email protected]:Jiahao-Ma/HashPoint-Code.git

4. Install hashpoint CUDA library

pip install .