bevfusion
bevfusion copied to clipboard
Published
20 hours ago •
mit-han-lab
mit-han-lab
KeyError: 'radar'`
I have an error when I try to use your code.
File "/home/user/Documents/GitHub/bevfusion/mmdet3d/datasets/pipelines/loading.py", line 728, in __call__ radars_dict = results['radar'] KeyError: 'radar'
And I found a simiar iusse in another project
Is it because I put the mmdetection3d github folder into the project and install it before I run python setup.py develop so the mmdetection3d could affect the result?
The following are the outputs after I run python setup.py develop
Removing mmdet3d 1.0.0rc0 from easy-install.pth file
Adding mmdet3d 0.0.0 to easy-install.pth file
detected new path '/home/cine/Documents/GitHub/bevfusion/mmdetection3d'
Installed /home/cine/Documents/GitHub/bevfusion
Processing dependencies for mmdet3d==0.0.0
Finished processing dependencies for mmdet3d==0.0.0
did you know how to solve this problem? Thank you very much for your help and reply~
Here I want to share what I have done with nuscene mini dataset and this code (even it is with some Chinese, I help this could help people want to use this code)
conda create --name fusionbev python=3.8 -y
pip install opencv-python
pip install pyquaternion
pip install nuscenes-devkit
pip install lyft-dataset-sdk
conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge
#下载openmpi 见最下面
pip install mpi4py==3.0.3
pip install torchpack
pip install tqdm
pip install Pillow==8.4.0
# mmcv = 1.4.0 mmdetection = 2.20.
# https://mmdetection.readthedocs.io/en/latest/get_started.html
pip install -U openmim # 安装 mim
mim install mmengine # 安装基础库 mmengine
# https://mmdetection.readthedocs.io/zh_CN/v2.21.0/get_started.html
### MMCV/MMDet/MMDet3D 的版本对应_csuzhaoqinghui的博客-CSDN博客
#### https://blog.csdn.net/csuzhaoqinghui/article/details/130984373
MMDetection3D MMDetection MMSegmentation MMCV
v1.0.0rc0 mmdet>=2.19.0, <=3.0.0 mmseg>=0.20.0, <=1.0.0 mmcv-full>=1.3.17, <=1.5.0
mim install "mmcv-full==1.4.0"
pip install mmdet==2.20.0
git clone https://github.com/open-mmlab/mmsegmentation.git
cd mmsegmentation
git checkout v0.20.0 # switch to v0.20.0 branch
pip install -e . # or "python setup.py develop"
cd ..
git clone https://github.com/open-mmlab/mmdetection3d.git
cd mmdetection3d
git checkout v1.0.0rc0 # switch to v0.20.0 branch
pip install -e . # or "python setup.py develop"
## 创建pkl文件
# https://mmdetection3d.readthedocs.io/zh_CN/dev-1.x/advanced_guides/datasets/nuscenes.html
# cd到mmdetection3d文件夹，注意root path多位置和version的设置
# ./tools/create_data.py line 120
# elif args.dataset == "nuscenes" and args.version == "v1.0-mini":
python ./tools/create_data.py nuscenes --root-path ../data/nuscenes --out-dir ../data/nuscenes --extra-tag nuscenes --version v1.0-mini
## mmdet3d 1.0.0rc0 requires numba==0.53.0, but you have numba 0.58.0 which is incompatible.
pip install numba==0.53.0
pip install numpy==1.23.4
# 执行完，增加文件
nuscenes_dbinfos_train.pkl
nuscenes_gt_database
nuscenes_infos_train_mono3d.coco.json
nuscenes_infos_train.pkl
nuscenes_infos_val_mono3d.coco.json
nuscenes_infos_val.pkl
#########################
git 主页的Evaluation权重需要手动点击链接下载
./tools/download_pretrained.sh # 这个命令行失效了，只能建立pretrain文件夹
cd .. # github项目目录
python setup.py develop
##################################################################
输出
##################################################################
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/spconv/sparse_conv_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/spconv
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/bev_pool/bev_pool_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/bev_pool
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/iou3d/iou3d_cuda.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/iou3d
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/voxel/voxel_layer.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/voxel
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/roiaware_pool3d/roiaware_pool3d_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/roiaware_pool3d
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/ball_query/ball_query_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/ball_query
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/knn/knn_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/knn
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/paconv/assign_score_withk_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/paconv
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/group_points/group_points_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/group_points
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/interpolate/interpolate_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/interpolate
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/furthest_point_sample/furthest_point_sample_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/furthest_point_sample
copying build/lib.linux-x86_64-cpython-38/mmdet3d/ops/gather_points/gather_points_ext.cpython-38-x86_64-linux-gnu.so -> mmdet3d/ops/gather_points
Creating /home/cine/miniconda3/envs/fusionbev/lib/python3.8/site-packages/mmdet3d.egg-link (link to .)
Removing mmdet3d 1.0.0rc0 from easy-install.pth file
Adding mmdet3d 0.0.0 to easy-install.pth file
detected new path '/home/cine/Documents/GitHub/bevfusion/mmdetection3d'
Installed /home/cine/Documents/GitHub/bevfusion
Processing dependencies for mmdet3d==0.0.0
Finished processing dependencies for mmdet3d==0.0.0
##################################################################
torchpack dist-run -np 8 python tools/test.py configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml pretrained/bevfusion-det.pth --eval bbox
torchpack dist-run -np 8 python tools/train.py configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth --load_from pretrained/lidar-only-det.pth
###################
#下载openmpi
# 参考 https://blog.csdn.net/qq_26822029/article/details/107930758
到openmpi官网下载openmpi，这里我使用的是4.0.4版本，也可以使用wget下载：
wget https://download.open-mpi.org/release/open-mpi/v4.0/openmpi-4.0.4.tar.gz
下载完成之后解压：
tar -zxvf openmpi-4.0.4.tar.gz
进入openmpi-4.0.4文件夹之后检查配置文件
cd openmpi-4.0.4
./configure
在漫长的等待之后编译安装
sudo make all install
编译完成之后需要配置一下openmpi的环境变量
vim /etc/profile
# 在这个文件末尾添加如下两行
export PATH=/usr/local/path:$PATH
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib
编辑完成之后需要执行source /etc/profile使其生效
进入到examples文件夹中，执行make编译一下测试代码，如果编译没有报错执行下面的测试语句
make # 先make再执行下面一句 ！！！！！
mpirun -np 4 hello_c
如果正常输出结果则说明安装完成~
@songhan @zhijian-liu @hopef @K-Shao I am sorry for this @ to you, could you please look at this issue and give me some hints? Thank you very much~~
