IFTR Installation Guide

Author: Shaohong Wang wangsh0111@zju.edu.cn

---

Installation

Create a conda environment

All operations should be done on machines with GPUs.

We expect a CUDA version displayed in the top right corner of the nvidia-smi to be higher than 11.6

conda create --name IFTR python=3.8 cmake=3.22.1
conda activate IFTR 

conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch

Install some other packages

cd IFTR-main
pip install -r requirements.txt

Install mmdet3d

pip install mmcv-full==1.4.0 -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.10.0/index.html
# recommend downloading .whl directly and then installing it
# pip install mmcv_full-1.4.0-cp38-cp38-manylinux1_x86_64.whl

pip install mmdet==2.14.0 mmsegmentation==0.14.1
pip install ninja tensorboard nuscenes-devkit==1.1.10 scikit-image==0.19.0 lyft-dataset-sdk
pip install numpy==1.19.5 pandas==1.4.4 scikit-image==0.19.3 llvmlite

git clone https://github.com/open-mmlab/mmdetection3d.git
cd mmdetection3d
git checkout v0.17.1 # Other versions may not be compatible.
python setup.py install

pip install einops fvcore seaborn iopath timm  typing-extensions pylint ipython numba yapf==0.40.1 scikit-image==0.19.3

python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'
# # Or, to install it from a local clone:
# git clone https://github.com/facebookresearch/detectron2.git
# cd detectron2
# python -m pip install -e ./
pip install numpy==1.19.5 pandas==1.4.4 scikit-image==0.19.3 nuscenes-devkit==1.1.10 

Install pypcd

git clone https://github.com/klintan/pypcd.git
cd pypcd
python setup.py install

Install IFTR

cd IFTR-main

python setup.py develop
# Bbx IOU cuda version compile
python opencood/utils/setup.py build_ext --inplace 

Troubleshooting

AttributeError: module 'distutils' has no attribute 'version'

pip uninstall setuptools                # cannot use conda uninstall setuptools
pip install setuptools==59.5.0

mmdetection error TypeError: FormatCode() got an unexpected keyword argument ‘verify‘

pip uninstall yapf
pip install yapf==0.40.1

libGL.so.1: cannot open shared object file: No such file or directory

apt-get update && apt-get install libgl1

Command

Suppose you are in the folder IFTR-main

Training

# Single GPU Training
CUDA_VISIBLE_DEVICES=0 python opencood/tools/train.py -y YAML_FILE [--model_dir MODEL_DIR]

# Multi-GPU training, nproc_per_node set to GPU number
CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 --use_env opencood/tools/train.py -y YAML_FILE [--model_dir MODEL_DIR]

• -y YAML_FILE the yaml configuration file
• [--model_dir MODEL_FOLDER] is optional, indicating that training continues from this log (resume training). It will read config.yaml from under MODEL_FOLDER instead of the input -y YAML_FILE. so it can be written -y None and no yaml file is provided

Testing

python opencood/tools/inference.py --model_dir MODEL_DIR [--fusion_method iftr]

• inference.py There are a number of optional parameters. Please go to the code to see details
• [--fusion_method iftr] The default fusion_method is iftr, which corresponds to the iftr model with the addition of late fusion strategy

Optional fusion_method include

• [--fusion_method intermediate] This corresponds to the naive iftr model

---