conda create -n capstone python=3.8 numpy cudatoolkit=10.2

Instalasi PyTorch dan Torchvision dapat mengikuti dua cara tergantung dari GPU yang digunakan. Untuk GPU terkini atau yang dipakai pada notebook dapat menginstall langsung melalui pip:

pip install torch==1.9.0 torchvision==0.10

Jika menggunakan GPU lama (Cuda Compute < 3.5), dapat mendownload .whl terlebih dahulu pada blog Nelson Liu. Setelahnya dapat melakukan installasi melalui pip:

pip install ./torch-1.9.0+cu102-cp38-cp38-linux_x86_64.whl
pip install ./torchvision-0.10.0+cu102-cp38-cp38-linux_x86_64.whl

Package/library lainnya yang diperlukan dapat diinstal melalui requirements.txt:

pip install -r requirements.txt

Clone DCNv2 for PyTorch > 1.8 from ruhyadi/DCNv2_18

cd src/lib/models/networks
rm -rf DCNv2
git clone https://github.com/ruhyadi/DCNv2_18 ./DCNv2

Build DCNv2 from scrach

cd DCNv2
python setup.py build develop

Clone IoU3D from ruhyadi/iou3d

cd src/lib/utils
rm -rf iou3d
git clone https://github.com/ruhyadi/iou3d ./iou3d

Build IoU3D from scract

cd iou3d
python setup.py install

Failed on PyTorch 1.9.0, but can inference model.

Docker image can be build with:

docker build -t username/rtm3d:latest .

Or you can pull docker image (recommend) with:

docker pull -t ruhyadi/rtm3d:latest

Then you can run docker container in interactive mode with:

./runDocker.sh

python src/tools/export_nuscenes.py

python src/tools/create_sets_nuscenes.py

python src/tools/nuscenes_to_coco.py

python ./src/faster.py \
    --demo ./demo_kitti_format/data/kitti/image \
    --calib_dir ./demo_kitti_format/data/kitti/calib/ \
    --load_model ./weights/model_res18_1.pth \
    --gpus 0 \
    --vis \
    --arch res_18