Skip to content

Latest commit

 

History

History
69 lines (47 loc) · 3 KB

SETUP.md

File metadata and controls

69 lines (47 loc) · 3 KB
Raw

Setup

The code was tested on Ubuntu 20.04/22.04 and CUDA 11.5/11.7, with Python 3.8.

1. Clone this repo recursively.

Our repo uses RAFT and DPT as out-of-the-box data processors for optical flow and monocular depth estimation, respectively. They are included as git submodules.

Note that for RAFT, we use our forked repo instead of the original one, where the only difference we made is that RAFT models are exposed as python modules for easy access.

# Make sure you include the `--recursive` arg.
git clone https://github.com/KAIR-BAIR/dycheck --recursive

2. Create a new conda environment.

This step assumes you have already installed anaconda. If you have not, please follow its instruction.

# Create a seperate python environment.
conda create -n dycheck python=3.8 -y

3. Install necessary dependencies.

Our default requirements are tailored to CUDA 11.5 and 11.7. If you use other CUDA versions, you might want to edit the requirement file by:

  1. Change the jax and jaxlib versions (instructions).
  2. Change the torch and torchvision versions (instructions).
# System level `tensorflow-graphics` dependency.
sudo apt install libopenexr-dev ffmpeg -y
# Main project dependencies.
pip install -r requirements.txt

Note that it will install the repo as an editable local package, where you can make any changes you want and they will be propagated into the environment automatically.

4. [Optional] Prepare for jupyterlab to process data.

For users who want to use our annotation tools to create and process their capture, please follow this step. You need to setup up jupyter environment and jupyter kernel.

# Setup labextensions for annotation and visualization.
jupyter labextension install jupyterlab-plotly ipyevents
# Make sure that you use the `python` executable from the project environment.
python -m ipykernel install --user --name=dycheck

5. [Optional] Prepare for rclone to pull datasets and checkpoints from google drive.

For users who want to use our provided scripts/instructions to download datasets and checkpoints, please follow this step. This step assumes you have already installed rclone, if you have not, please follow its instruction.

# Set up `rclone` with your google drive. You only need to run this once.
rclone config

Then click on our released folder to open it in your google drive. You can verify by running the following command to make sure that it is accessible through terminal:

# `<DRIVE_NAME>` is the name of your google drive profile for `rclone`.
rclone lsd --drive-shared-with-me "<DRIVE_NAME>:/dycheck-release"