Mask RCNN(Pytorch)的使用安装

GitHub - facebookresearch/maskrcnn-benchmark: Fast, modular reference implementation of Instance Segmentation and Object Detection algorithms in PyTorch.

安装时一直提示报错fatal error: THC/THC.h: No such file or directory

主要原因是pytorch版本过高要选择1.10的，1.11的就会过高[Debug] Pytorch 版本问题 THC/THC.h: No such file or directory_文三路张同学的博客-CSDN博客

conda create -n maskrcnn python=3.7
conda activate maskrcnn

# pytorch官网下载1.10版本

pip install torch==1.10.1+cu111 torchvision==0.11.2+cu111 torchaudio==0.10.1 -f https://download.pytorch.org/whl/cu111/torch_stable.html

conda install ipython pip
pip install ninja yacs cython matplotlib tqdm opencv-python -i https://pypi.tuna.tsinghua.edu.cn/simple

# 安装pycoco
git clone https://github.com/cocodataset/cocoapi.git
cd cocoapi/PythonAPI
python setup.py build_ext install


# 安装apex
git clone https://github.com/NVIDIA/apex
cd apex
python setup.py install

#安装maskrcnn
cd maskrcnn-benchmark
python setup.py install

然后会出现error: ‘AT_CHECK’ was not declared in this scope解决方法（版本过高的问题但是30系列的需要11.0以上的，因此pytorch最好选这个）

解决方法：将所有待编译的源文件中的‘AT_CHECK’全部替换为‘TORCH_CHECK’。

error: ‘AT_CHECK’ was not declared in this scope解决方法_hailanyi的博客-CSDN博客

安装好了后执行测试

cd demo

python webcam.py --min-image-size 800

AttributeError:模块'torch._six‘没有特性'PY3’

https://www.cnblogs.com/BlogLwc/p/15663776.html

查看安装时的错误提示修改imports.py和c2_model_loading.py里的PY3成PY37(python版本)

Can‘t parse ‘org‘. Sequence item with index 0 has a wrong type

原因是opencv与python版本不匹配

去清华镜像Links for opencv-python (tsinghua.edu.cn)

找cp37的whl文件下载（因为之前已经安装了opencv-pytorch的记得先删除）

pip uninstall opencv-python

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple opencv-python==3.4.10.35

Can‘t parse ‘org‘. Sequence item with index 0 has a wrong type_lolitaaaa的博客-CSDN博客

最后就安装成功了