How do you build this?

[juergenmathes]

I'm wondering how you build this repo. I think the dockerfile is outdated and the mmdetection3d is on a way newer state.

Any advice? -> My goal is to create a docker image. ... If nobody answers I will just experiment until it works.

[XuyangBai]

Hi, Yes I build this project on an early version of mmdet3d. You can create the docker image by something like:

ARG PYTORCH="1.6.0"
ARG CUDA="10.1"
ARG CUDNN="7"

FROM pytorch/pytorch:${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel

ENV TORCH_CUDA_ARCH_LIST="6.0 6.1 7.0+PTX"
ENV TORCH_NVCC_FLAGS="-Xfatbin -compress-all"
ENV CMAKE_PREFIX_PATH="$(dirname $(which conda))/../"

RUN apt-get update && apt-get install -y ffmpeg libsm6 libxext6 git ninja-build libglib2.0-0 libsm6 libxrender-dev libxext6 \
    && apt-get clean \
    && rm -rf /var/lib/apt/lists/*

# Install MMCV
RUN pip install mmcv-full==latest+torch1.6.0+cu101 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html
RUN pip install mmdet

# Install MMDetection
RUN conda clean --all
RUN git clone https://github.com/XuyangBai/TransFusion.git /TransFusion
WORKDIR /TransFusion
ENV FORCE_CUDA="1"
RUN pip install -r requirements/build.txt
RUN pip install --no-cache-dir -e .

# uninstall pycocotools installed by nuscenes-devkit and reinstall mmpycocotools
RUN pip uninstall pycocotools --no-cache-dir -y
RUN pip install mmpycocotools --no-cache-dir --force --no-deps

Basically just need to built TransFusion the same way as you did for mmdet3d.