Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
Zmjcc 447436b8e8 | 2 years ago | |
---|---|---|
.. | ||
origin_tools | 2 years ago | |
LICENSE | 2 years ago | |
Makefile | 2 years ago | |
README.md | 2 years ago | |
requirements.txt | 2 years ago | |
setup.py | 2 years ago | |
wireless_channel.py | 2 years ago |
PCDet
is a general PyTorch-based codebase for 3D object detection from point cloud.
PCDet
is a general PyTorch-based codebase for 3D object detection from point cloud.
It currently supports several state-of-the-art 3D object detection methods (PointPillar
, SECOND
, Part-A^2 Net
) with highly refactored codes for both one-stage and two-stage frameworks.
This is also the official code release of Part-A^2 net
.
Note that currently this framework mainly contains the voxel-based approaches and we are going to support more point-based approaches in the future.
Supported methods are shown in the below table. The results are the 3D detection performance of car class on the val set of KITTI dataset.
All models are trained with 8 GPUs and are available for download.
training time | AP_Easy | AP_Mod. | AP_Hard | download | |
---|---|---|---|---|---|
PointPillar | ~2hours | 87.37 | 77.30 | 74.02 | model-18M |
SECOND | ~2hours | 88.46 | 78.46 | 76.63 | model-20M |
Part-A^2 | ~5hours | 89.66 | 79.45 | 78.80 | model-209M |
Part-A^2-fc | ~5hours | 89.57 | 79.31 | 78.61 | model-244M |
All the codes are tested in the following environment:
pcdet
git clone https://github.com/sshaoshuai/PCDet.git
pip install -r requirements.txt
spconv
. Note that we use an old version of spconv
, make sure you install the spconv v1.0
(commit 8da6f96) instead of the latest one.pcdet
library by running the following command:python setup.py develop
Currently we only support KITTI dataset, and contributions are welcomed to support more datasets.
PCDet
├── data
│ ├── kitti
│ │ │──ImageSets
│ │ │──training
│ │ │ ├──calib & velodyne & label_2 & image_2 & (optional: planes)
│ │ │──testing
│ │ │ ├──calib & velodyne & image_2
├── pcdet
├── tools
pcdet/datasets/kitti
:python kitti_dataset.py create_kitti_infos
All the config files are within tools/cfgs/
.
python test.py --cfg_file ${CONFIG_FILE} --batch_size 4 --ckpt ${CKPT}
The '--ckpt' will load param file from
model
folder, by default.
Part-A^2
model, please run the following command (here we add --set
to modify some default parameters to match with the training setting of the provided Part-A^2
model, and other provided models do not need to add it):python test.py --cfg_file cfgs/PartA2_car.yaml --batch_size 4 --ckpt PartA2.pth \
--set MODEL.RPN.BACKBONE.NAME UNetV0 MODEL.RPN.RPN_HEAD.ARGS use_binary_dir_classifier:True
--eval_all
argument:python test.py --cfg_file ${CONFIG_FILE} --batch_size 4 --eval_all
Currently, to train PointPillar
or SECOND
or PartA2
, the --batch_size
depends on the number of your training GPUs as we use ${BATCH_SIZE}=4*${NUM_GPUS}
, i.e., --batch_size 32
for training with 8 GPUs.
sh scripts/dist_train.sh ${NUM_GPUS} \
--cfg_file ${CONFIG_FILE} --batch_size ${BATCH_SIZE}
sh scripts/slurm_train.sh ${PARTITION} ${JOB_NAME} ${NUM_GPUS} \
--cfg_file ${CONFIG_FILE} --batch_size ${BATCH_SIZE}
python train.py --cfg_file ${CONFIG_FILE} --batch_size ${BATCH_SIZE}
We would like to thank for second.pytorch for providing the original implementation of the one-stage voxel-based framework,
and there are also some parts of the codes that are modified from PointRCNN.
We hope that this repo could serve as a strong and flexible codebase to benefit the research community by speeding up the process of reimplementing previous works and/or developing new methods.
If you find this work useful in your research, please consider cite:
@article{shi2020points,
title={From Points to Parts: 3D Object Detection from Point Cloud with Part-aware and Part-aggregation Network},
author={Shi, Shaoshuai and Wang, Zhe and Shi, Jianping and Wang, Xiaogang and Li, Hongsheng},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2020},
publisher={IEEE}
}
and / or
@inproceedings{shi2019pointrcnn,
title={PointRCNN: 3d Object Progposal Generation and Detection from Point Cloud},
author={Shi, Shaoshuai and Wang, Xiaogang and Li, Hongsheng},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={770--779},
year={2019}
}
Should you have any question, please contact Shaoshuai Shi (@sshaoshuai).
Net4AI 是未来(6G)内生智慧网络的两大技术发展方向之一,该方向的研究目标为充分利用网络节点的通信、计算和感知能力,通过分布式学习、群智式协同以及云边端一体化算法部署,支撑更为强大的网络智能(Network Intelligence),以实现未来各种智慧应用。本项目将对Net4AI相关论文的实现代码进行汇总,方便相关论文结果的复现,以促进该领域的健康发展。
Unity3D Asset CSV Text Python MATLAB other
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》