Skip to content

Latest commit

 

History

History
112 lines (91 loc) · 2.95 KB

memo.md

File metadata and controls

112 lines (91 loc) · 2.95 KB
Raw

手順メモ

環境初期化

python3 -m venv .env
source .env/bin/activate
pip3 install torch torchvision
git clone https://github.com/SakodaShintaro/street-gaussians-ns
git submodule update --init --recursive
pip3 install rawpy==0.21.0
pip3 install numpy==1.24.3 pandas==2.2.1
pip3 install ninja git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch
pip3 install nerfstudio
pip3 install "git+https://github.com/facebookresearch/pytorch3d.git"
pip3 install -e .
pip3 install dependencies/nvdiffrast/
cd dependencies/detectron2
pip3 install -e .
cd ../Mask2Former
pip3 install -r requirements.txt
cd mask2former/modeling/pixel_decoder/ops
sh make.sh
sudo apt install -y ffmpeg

pip3 install --upgrade pip
pip3 install waymo-open-dataset-tf-2-12-0

colmap

aptで入れたもの COLMAP 3.7 (Commit Unknown on Unknown without CUDA) では colmap model_comparer というコマンドが存在しなかった。

ソースコードから3.9.1をビルドする。 https://colmap.github.io/install.html

sudo apt-get remove --purge colmap
git clone https://github.com/colmap/colmap.git
cd colmap
git checkout tags/3.9.1
mkdir build
cd build
<CMakeLists.txtに set(CMAKE_CUDA_ARCHITECTURES "native")を追加>
cmake .. -GNinja
ninja
sudo ninja install

学習実行

python3 ./scripts/pythons/extract_waymo.py --waymo_root ./data/ --out_root ./data/extracted
bash scripts/shells/data_process.sh ./data/extracted
bash scripts/shells/train.sh ./data/extracted 0

Dockerを使って環境構築する場合

nerf-studioのドキュメントを参考にする

https://docs.nerf.studio/quickstart/installation.html#use-docker-image

https://hub.docker.com/r/dromni/nerfstudio

から最新を取得する。現在だと

docker run --gpus all \
            -v $HOME/data:/home/user/data/ \
            -v $HOME/.cache/:/home/user/.cache/ \
            -v /media:/media \
            -p 7007:7007 \
            -it \
            --ipc=host \
            --privileged \
            my-pytorch-image \
            /bin/bash
sudo apt install -y ffmpeg libgl1-mesa-glx libglib2.0-0
git submodule update --init --recursive
pip3 install "git+https://github.com/facebookresearch/pytorch3d.git"
pip3 install -e .
pip3 install dependencies/nvdiffrast/
cd dependencies/detectron2
pip3 install -e .
cd ../Mask2Former
pip3 install -r requirements.txt
cd mask2former/modeling/pixel_decoder/ops
sh make.sh
# これは権限の問題で上手くいかない
# make.shスクリプトの末尾に--userをつければいい
python3 setup.py build install --user

sudo apt install -y libgl1-mesa-glx libglib2.0-0
sudo apt install tzdata

イメージをDockerfileから作る

docker build -t my-pytorch-image .

モデルのコピー

wget -P dependencies/Mask2Former/models/ https://dl.fbaipublicfiles.com/maskformer/mask2former/mapillary_vistas/semantic/maskformer2_swin_large_IN21k_384_bs16_300k/model_final_90ee2d.pkl