用docker部署SparseInst项目踩坑

论文：Sparse Instance Activation for Real-Time Instance Segmentation

原本官方提供dockerfile文件为：

FROM pytorch/pytorch:1.11.0-cuda11.3-cudnn8-devel
LABEL Service="SparseInstanceActivation"

ENV TZ=Europe/Moscow
ENV DETECTRON_TAG=v0.3
ARG DEBIAN_FRONTEND=noninteractive

RUN apt-key del 7fa2af80 && \
    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
RUN apt update && apt install vim git g++ python3-tk ffmpeg libsm6 libxext6 -y

RUN python3 -m pip install --no-cache-dir --upgrade pip && \
    python3 -m pip install --no-cache-dir opencv-python opencv-contrib-python scipy

WORKDIR /workspace
RUN git clone https://github.com/facebookresearch/detectron2.git && \
    cd detectron2/ && git checkout tags/${DETECTRON_TAG} && python3 setup.py build develop

RUN python3 -m pip uninstall -y iopath fvcore portalocker yacs && \
    python3 -m pip install --no-cache-dir iopath fvcore portalocker yacs timm pyyaml==5.1 shapely

RUN git clone https://github.com/hustvl/SparseInst
WORKDIR /workspace/SparseInst
RUN ln -s /usr/bin/python3 /usr/bin/python

ENTRYPOINT bash

执行

docker build -t my_custom_image .

1.执行后报错

gcc: error: pycocotools/_mask.c: No such file or directory

查询后得知是没安装cython导致，在dockerfile文件里加上安装cython。

2.报错变为

FutureWarning: Cython directive 'language_level' not set, using '3str' for now (Py3)

提前安装pycocotools=2.0.8解决

3.然后遇到报错

Collecting pyyaml==5.1
Downloading PyYAML-5.1.tar.gz (274 kB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 274.2/274.2 kB 5.0 MB/s eta 0:00:00
error: subprocess-exited-with-error

安装pyyaml==6.0.1版本可正确执行

4.进入docker环境后执行以下指令报错，说detectron2不存在

python tools/test_net.py --config-file configs/sparse_inst_r50_giam_fp16.yaml --fp16 MODEL.WEIGHTS model_final.pth 

 去detectron2官方找到了新的下载安装方式，写入dockerfile文件。发现还是不行，需要进入容器之后执行

git clone https://github.com/facebookresearch/detectron2.git

这样就能顺利了

5.将coco数据集放在正确路径之后，报错RuntimeError: nvrtc: error: invalid value for --gpu-architecture (-arch)

原因：torch版本不对，应该将torch升级到13.0或者以上的版本，我升级到了13.1

6.numpy版本2.1太高

更改numpy版本为1.23.5

解决后的dockerfile文件：

FROM pytorch/pytorch:1.13.1-cuda11.6-cudnn8-devel
LABEL Service="SparseInstanceActivation"

ENV TZ=Europe/Moscow
ENV DETECTRON_TAG=v0.6
ARG DEBIAN_FRONTEND=noninteractive

RUN apt-key del 7fa2af80 && \
    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
RUN apt update && apt install vim git g++ python3-tk ffmpeg libsm6 libxext6 -y

RUN python3 -m pip install --no-cache-dir --upgrade pip && \
    python3 -m pip install --no-cache-dir opencv-python opencv-contrib-python scipy cython 

WORKDIR /workspace
RUN  git clone https://github.com/facebookresearch/detectron2.git && python3 -m pip install -e detectron2

RUN python3 -m pip uninstall -y iopath fvcore portalocker yacs && \
    python3 -m pip install --no-cache-dir iopath fvcore portalocker yacs timm pyyaml==6.0.1 shapely
RUN python3 -m pip install --no-cache-dir numpy==1.23.5

RUN git clone https://github.com/hustvl/SparseInst
WORKDIR /workspace/SparseInst
RUN ln -s /usr/bin/python3 /usr/bin/python

ENTRYPOINT bash