Skip to content

lturing/qd_3dt_modified

Repository files navigation

运行(环境ubuntu20)

update at 2023-12-10
add MoMA-M3T

单目2d和3d目标检测和跟踪以及bev可视化。qd-3dt相比常见的目标检测器,引入基于lstm的运动模型实现3d目标跟踪,同时实现跨帧间目标的信息共享,进一步优化目标检测器的结果,使得检测更鲁棒。
参考官方代码的基础下做了以下调整:

  • 简化代码,保留核心部分
  • 支持任意分辨率的图像输入
  • 只支持cpu(gpu需要调整部分代码)
  • 添加bev可视化

值得注意的是,该模型的泛化性不一定好,因而使用自己的数据集前,需要重新训练,请参照官方的教程。运行bev模式时,需要提供Twc,若数据没Twc,可以通过slam等获取。

# 安装pypanglin
git clone -b v0.8 --recursive https://github.com/stevenlovegrove/Pangolin.git
cd Pangolin

# See what package manager and packages are recommended
./scripts/install_prerequisites.sh --dry-run recommended

# Install dependencies (as described above, or your preferred method)
./scripts/install_prerequisites.sh recommended

# Configure and build
cmake -B build -DPython_EXECUTABLE=`which python3` 
# cmake -B build -DCMAKE_INSTALL_PREFIX=/home/spurs/installed -DPython_EXECUTABLE=`which python3`
cmake --build build -t pypangolin_pip_install
# uninstall 
# cmake --build build -t pypangolin_pip_uninstall

# 模型参数下载
百度链接(kitti): https://pan.baidu.com/s/1-ElUcDVL-YOOxrLCgdDmBA 提取码: bt3g

官方链接:https://github.com/SysCV/qd-3dt/blob/main/readme/MODEL_ZOO.md

# kitti测试数据下载
链接:https://s3.eu-central-1.amazonaws.com/avg-kitti/raw_data/2011_10_03_drive_0047/2011_10_03_drive_0047_sync.zip
相关参数:https://s3.eu-central-1.amazonaws.com/avg-kitti/raw_data/2011_10_03_calib.zip

更多的kitti数据见https://zhuanlan.zhihu.com/p/664386718

# 下载代码
git clone https://github.com/lturing/qd_3dt_modified

# 运行
cd qd_3dt_modified
python run_kitti.py --config quasi_dla34_dcn_3dmatch_multibranch_conv_dep_dim_cen_clsrot_sep_aug_confidence_mod_anchor_ratio_small_strides_GTA.py --checkpoint path_to_latest_kitti.path --lstm_checkpoint path_to_batch8_min10_seq10_dim7_train_dla34_regress_pretrain_VeloLSTM_kitti_100_linear.pth --data path_to_kittii_data --pose path_to_kitti_data_pose -- cali path_to_kitti_cali 

demo(b站)

感谢

About

单目2d和3d目标检测,bev可视化

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages