Update docker file

[ranran9991]

🚀 Feature

Fix the docker file

I've been struggling endlessly trying to make by tuning the versions of pytorch, cuda, and torchvision and it just wont work.

[KaihuaTang]

if you want to use docker, pls go directly check https://github.com/facebookresearch/maskrcnn-benchmark My code is based on their project.

[ranran9991]

I've tried adapting and changing the docker file to fit it to your project but I either get stuck with installing apex or get stuck with cryptic and weird errors when running the training, depending on the versions of python and cuda (I tried python 3.6 and 3.7, cuda 9.0, 10.0, 10.1, 10.2).

I would really love if you could try to make it work.

[roshambo919]

FROM pytorch/pytorch:1.8.0-cuda11.1-cudnn8-devel

RUN echo 'debconf debconf/frontend select Noninteractive' | debconf-set-selections

# To fix GPG key error when running apt-get update
RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/3bf863cc.pub && \
    apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/7fa2af80.pub

# install basics
RUN apt-get update -y \
 && apt-get install -y apt-utils vim nano git curl ca-certificates bzip2 cmake tree htop bmon iotop g++ \
 && apt-get install -y libglib2.0-0 libsm6 libxext6 libxrender-dev ffmpeg libsm6 libxext6 libqt5gui5

# Create a Python environment
RUN conda install -y conda-build ipython \
 && conda clean -ya \
 && pip install requests ninja yacs cython matplotlib opencv-python tqdm h5py

# Install TorchVision master
RUN git clone https://github.com/pytorch/vision.git \
 && cd vision \
 && git fetch --all --tags --force \
 && git checkout tags/v0.9.0 -b v0.9.0-branch \
 && python setup.py install

# install pycocotools
RUN git clone  https://github.com/ppwwyyxx/cocoapi.git \
 && cd cocoapi/PythonAPI \
 && python setup.py build_ext install

# install apex
RUN git clone https://github.com/NVIDIA/apex.git \
 && cd apex \
 && git fetch --all --tags --force \
 && git checkout tags/22.03 -b v22.03-branch \
 && python setup.py install --cuda_ext --cpp_ext

# install PyTorch Detection
ARG FORCE_CUDA="1"
ENV FORCE_CUDA=${FORCE_CUDA}
ENV TORCH_CUDA_ARCH_LIST="6.0 6.1 7.0+PTX"
ENV TORCH_NVCC_FLAGS="-Xfatbin -compress-all"
ENV CMAKE_PREFIX_PATH="$(dirname $(which conda))/../"
COPY . scenegraph
RUN cd scenegraph \
 && pip install -r requirements.txt scipy \
 && python setup.py build develop

This is working for me so far. Note that you'll have to update some deprecated features in the code such as the use of np.float and np.bool, among others. I can post a link to my fork if desired.