openpcdet-LION环境配置

简要介绍

ToDo

• [ ]

---

Python 3.8+torch 2.1+cuda 11.8

pip install  --timeout 300 \
    --retries 3 \
    -i https://pypi.tuna.tsinghua.edu.cn/simple \
    torch==2.1.0 \
    torchvision==0.16.0 \
    torchaudio==2.1.0 \
    --index-url https://download.pytorch.org/whl/cu121

稀疏卷积

pip install cumm-cu118  spconv-cu118

Mamba 2.2

对于旧版本一定要下载 mamba 2.2，torch>=2.4 时安装 2.3 版本

# 推荐安装组合
pip install causal-conv1d==1.4.0 mamba-ssm==2.2.2 -i https://pypi.tuna.tsinghua.edu.cn/simple

pip install  --no-build-isolation  mamba-ssm[causal-conv1d]

1.5

pip install --no-binary :all: causal-conv1d

Openpcdet

项目文件夹内

pip install -e .

pip install torch_scatter
pip install timm

pip install "opencv-python<4.12"

pip install nvidia-ml-py

1. 确保步幅为8的倍数

确保 d_model * expand / headdim 是8的倍数。例如，如果 d_model=192，expand=2，则 headdim 应该设置为48。

示例：

model = Mamba2(

d_model=192,

d_state=64,

d_conv=4,

expand=2,

headdim=48,

).to("cuda")

安装 openpcdet

pip install -e . -i https://mirrors.aliyun.com/pypi/simple/