yup. in this case it wasn't root. Removing that USER and -u in pip solves the problem. However, in our production images, we are required to remove root access.
` FROM nvidia/cuda:10.1-cudnn7-devel

ENV DEBIAN_FRONTEND noninteractive
RUN apt-get update && apt-get install -y
python3-opencv ca-certificates python3-dev git wget sudo ninja-build
RUN ln -sv /usr/bin/python3 /usr/bin/python

create a non-root user

ARG USER_ID=1000
RUN useradd -m --no-log-init --system --uid ${USER_ID} appuser -g sudo
RUN echo '%sudo ALL=(ALL) NOPASSWD:ALL' >> /etc/sudoers
USER appuser
WORKDIR /home/appuser

ENV PATH="/home/appuser/.local/bin:${PATH}"
RUN wget &&
python3 get-pip.py --user &&
rm get-pip.py

install dependencies

See for other options if you use a different version of CUDA

RUN pip install --user tensorboard cmake # cmake from apt-get is too old
RUN pip install --user torch==1.8 torchvision==0.9 -f

RUN pip install --user 'git+ '

install detectron2

RUN git clone detectron2_repo

set FORCE_CUDA because during docker build cuda is not accessible

ENV FORCE_CUDA="1"

This will by default build detectron2 for all common cuda architectures and take a lot more time,

because inside docker build, there is no way to tell which architecture will be used.

ARG TORCH_CUDA_ARCH_LIST="Kepler;Kepler+Tesla;Maxwell;Maxwell+Tegra;Pascal;Volta;Turing"
ENV TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST}"

RUN pip install --user -e detectron2_repo

Set a fixed model cache directory.

ENV FVCORE_CACHE="/tmp"
WORKDIR /home/appuser/detectron2_repo

run detectron2 under user "appuser":

wget -O input.jpg

python3 demo/demo.py \

#--config-file configs/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml \
#--input input.jpg --output outputs/ \
#--opts MODEL.WEIGHTS  `