Skip to content

Latest commit

 

History

History
55 lines (37 loc) · 2.81 KB

README.md

File metadata and controls

55 lines (37 loc) · 2.81 KB
Raw

Xihe Server Code

Installation

This repo is developed with Python, CUDA C++, and bash scripts. Please use the following commands to install the python dependencies first.

pipenv shell
pipenv install

pip install torch==1.8.0+cu111 torchvision==0.9.0+cu111 torchaudio==0.8.0 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch-cluster -f https://pytorch-geometric.com/whl/torch-1.8.0+cu111.html --upgrade

Dataset and Model

One of the key steps in reproducing our work is to generate the transformed point cloud datasets. For simplicity, we provide a pre-generated testing dataset at the data/dataset/ directory.

Please unzip the data/dataset/pointar.zip, data/dataset/xihe.zip and data/dataset/xihe_fps.zip files, and change the corresponding path definition in config.py.

Our pre-trained XiheNet model can be found at server/model/model.ckpt. To train the XiheNet from scratch, you will need to have access to the training dataset. Due to the large size of the training dataset (~TBs), it is difficult to provide directly---instead we provide the instructions for you to generate the dataset below.

Start the Service

Use the following code to start the Xihe server:

./launch.py serve

Directory Structure

  • datasets: datasets definitions and loaders.
  • model: XiheNet model definition and checkpoints
  • service: Xihe backend service
  • utils3d: essential 3D transformation and utilities for 3D math operations.
  • configs.py: path definition for datasets.
  • launch.py: command line entry script
  • Pipfile: pipenv dependency definition file

Generating Training Dataset

(The instructions below are adapted from the one provided in PointAR.)

We provide the scripts to generate their respective datasets. At the high level,

  • users should first obtain access to the two open-source datasets (i.e., Matterport3D and Neural Illumination datasets);
  • download these two datasets to a desirable directory. For the Matterport3D dataset, unzip the downloaded zip files and place them in a directory with structure similar to v1/scans/<SCENE_ID>/.... For the Neural Illumination dataset, just store the downloaded zip files, i.e. illummaps_<SCENE_ID>.zip, directly in a directory.
  • modify the corresponding the path variable in config.py file to reflect the local directory name;
  • then run the following commands to start generation.
    • PointAR: ./launch.py gen_pointar_data --dataset=test/train
    • Xihe: ./launch.py gen_xihe_data --dataset=test/train
    • XiheFPS: ./launch.py gen_xihe_fps_data --dataset=test/train

Note it can take a few hours to generate the entire dataset (~1.4TB) depending on the GPU devices.