DytanVO 代码复现(服务器端复现rtx3090)

源码地址

代码地址：https://github.com/castacks/DytanVO

环境配置

1.克隆github项目：

git clone https://github.com/castacks/DytanVO.git

2.利用yaml创建conda 环境：

修改yaml文件

name: dytanvo
channels:
  - pytorch
  - conda-forge
dependencies:
  - python=3.8
  - numba
  - tqdm
  - tbb
  - joblib
  - h5py
  - pytorch=1.7.0
  - torchvision=0.8.0
  - cudatoolkit=11.0
  - pip
  - toml=0.10.2
  - tomli=2.0.1
  - kornia=0.5.3

cd DytanVO
conda env create -f environment.yml
conda activate dytanvo

3.创建一个requirements.txt,安装相关的库

    absl-py==0.11.0
    antlr4-python3-runtime==4.9.3
    appdirs==1.4.4
    beautifulsoup4==4.11.1
    black==21.4b2
    cachetools==4.1.1
    chardet==3.0.4
    charset-normalizer==2.1.1
    cloudpickle==1.6.0
    cupy-cuda110
    cython==0.29.21
    data==0.4
    dataclasses==0.6
    # dcnv2==0.1
    decorator==5.1.1
    fastrlock==0.8
    filelock==3.8.0
    funcsigs==1.0.2
    future==0.18.2
    fvcore==0.1.2.post20201122
    gdown==4.5.1
    google-auth==1.23.0
    google-auth-oauthlib==0.4.2
    grpcio==1.34.0
    hydra-core==1.2.0
    idna==2.10
    imageio==2.9.0
    importlib-resources==5.9.0
    iopath==0.1.8
    joblib==0.17.0
    jsonpatch==1.32
    jsonpointer==2.3
    latex==0.7.0
    lxml==4.9.1
    markdown==3.3.3
    mypy-extensions==0.4.3
    # ngransac==0.0.0
    numpy==1.23.2
    oauthlib==3.1.0
    omegaconf==2.2.3
    opencv-python==4.4.0.46
    packaging==21.3
    pathspec==0.10.1
    portalocker==2.0.0
    protobuf==3.14.0
    pyasn1==0.4.8
    pyasn1-modules==0.2.8
    # pycocotools==2.0.4
    pydot==1.4.1
    pypng==0.0.20
    pysocks==1.7.1
    pytransform3d==1.14.0
    pyzmq==23.2.1
    regex==2022.8.17
    requests==2.25.0
    requests-oauthlib==1.3.0
    rsa==4.6
    shutilwhich==1.1.0
    soupsieve==2.3.2.post1
    splines==0.2.0
    tabulate==0.8.7
    tempdir==0.7.1
    tensorboard==2.4.0
    tensorboard-data-server==0.6.1
    tensorboard-plugin-wit==1.7.0
    timm==0.6.7
    toml==0.10.2
    torchfile==0.1.0
    tqdm==4.54.0
    trimesh==3.9.3
    urllib3==1.26.2
    visdom==0.1.8.9
    websocket-client==1.4.0
    werkzeug==1.0.1
    workflow==1.0
    zipp==3.8.1

pip install -r requirements.txt

4.编译DCNv2

cd Network/rigidmask/networks/DCNv2/; 
python setup.py install; 
cd -

下载模型和数据集

根据github的链接来下载DynaKITTI
https://drive.google.com/file/d/1BDnraRWzNf938UsfprWIkcqCSfOUyGt9/view

(另外一个数据集AirDOS-Shibuya给的链接没办法下载)

下载后解压到对应文件夹

运行

创建一个放结果的文件夹

mkdir results

创建一个run.sh的脚本，在脚本里输入（修改了一下模型名称）

traj=00_1
python -W ignore::UserWarning vo_trajectory_from_folder.py --vo-model-name vonet.pkl  \
							   --seg-model-name segnet-kitti.pth  \
							   --kitti --kitti-intrinsics-file data/DynaKITTI/$traj/calib.txt  \
							   --test-dir data/DynaKITTI/$traj/image_2  \
							   --pose-file data/DynaKITTI/$traj/pose_left.txt 

运行脚本

bash run.sh

跑起来了，不容易呀，复现了这么久

AirDOs

traj=RoadCrossing03
python -W ignore::UserWarning vo_trajectory_from_folder.py --flow-model-name flownet.pkl  \
							   --pose-model-name posenet.pkl  \
							   --seg-model segnet-sf.pth  \
							   --airdos  \
							   --test-dir data/AirDOS_shibuya/$traj/image_0  \
							   --pose-file data/AirDOS_shibuya/$traj/gt_pose.txt 

SceFLow跑好像要出问题
真值的格式不一样，不能直接用
–save-flow经测试其实并不行，代码写的有问题

img=Driving/frames_finalpass/15mm_focallength/scene_forwards/fast/left 
pose=Driving/camera_data/15mm_focallength/scene_forwards/fast/camera_data.txt 
python -W ignore::UserWarning vo_trajectory_from_folder.py --flow-model-name flownet.pkl  \
							   --pose-model-name posenet.pkl  \
							   --seg-model segnet-sf.pth  \
							   --sceneflow  \
							   --test-dir data/SceneFlow/$img  \
							   --pose-file data/SceneFlow/$pose 
							#    --save-flow \
							#    --batch-size 10 \
							#    --worker-num 5