Skip to content

install.md

Latest commit

 

History

History
56 lines (46 loc) · 2.2 KB

install.md

File metadata and controls

56 lines (46 loc) · 2.2 KB
Raw

Data and Environment Preparation

This guide installs the FlowR2A dependencies and downloads the data and models needed to run evaluation. FlowR2A is built on the NAVSIM devkit, so the data layout and environment variables follow the NAVSIM convention.

1. Clone the repository

git clone https://github.com/lixirui142/FlowR2A.git
cd FlowR2A

2. Download the data

Evaluation runs on the NAVSIM navtest split, which requires the nuPlan maps and the OpenScene test split. Download scripts are provided in download/.

NOTE: Please check the nuPlan LICENSE before downloading the data.

cd download
bash download_maps.sh
bash download_test.sh
cd ..

Organize the downloaded data exactly as NAVSIM specifies (see the NAVSIM install doc):

<WORKSPACE>/navsim_workspace
├── FlowR2A          # this repo (the devkit)
├── exp              # experiment / cache outputs
└── dataset
    ├── maps
    ├── navsim_logs
    │   └── test
    └── sensor_blobs
        └── test

Set the required environment variables (add to your ~/.bashrc). Replace <WORKSPACE> with any directory you prefer:

export WORKSPACE="<WORKSPACE>"   # e.g. /path/to/your/workspace
export NUPLAN_MAP_VERSION="nuplan-maps-v1.0"
export NUPLAN_MAPS_ROOT="$WORKSPACE/navsim_workspace/dataset/maps"
export NAVSIM_EXP_ROOT="$WORKSPACE/navsim_workspace/exp"
export NAVSIM_DEVKIT_ROOT="$WORKSPACE/navsim_workspace/FlowR2A"
export OPENSCENE_DATA_ROOT="$WORKSPACE/navsim_workspace/dataset"

3. Install the environment

Create the conda environment and install FlowR2A in editable mode:

conda env create --name flowr2a -f environment.yml
conda activate flowr2a
pip install -e .

Reminder: FlowR2A is developed and tested with torch==2.1.0 / torchvision==0.16.0 (pinned in requirements.txt). If this does not match your CUDA version, install the matching PyTorch build from pytorch.org.

Next: Evaluation.