安装detectron2

Tiu MoAbout 2 min

环境要求

Linux or macOS with Python ≥ 3.6
PyTorch ≥ 1.6 and torchvisionopen in new window that matches the PyTorch installation. Install them together at pytorch.orgopen in new window to make sure of this
OpenCV is optional but needed by demo and visualization
gcc & g++ ≥ 5.4 are required. ninjaopen in new window is optional but recommended for faster build.

安装detectron2

'Check the vision of pytorch' -> https://pytorch.org/get-started/locally/

# install pytorch
$ conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch-lts -c nvidia

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch -c nvidia

# install cv2 
$  'if xcb error' $ pip3 install opencv-python-headless 

# git project
$ git clone https://github.com/facebookresearch/detectron2.git
# Install Pre-Built Detectron2 (Linux only)
$ python -m pip install detectron2 -f \
  https://dl.fbaipublicfiles.com/detectron2/wheels/cu111/torch1.8/index.html

# Build Detectron2 from Source
$ python -m pip install -e detectron2

'To rebuild detectron2 that is built from a local clone, use' $rm -rf build/ **/*.so 'to clean the old build first. You often need to rebuild detectron2 after reinstalling PyTorch.'

opencv-python 的区别
opencv-python：这个存储库只包含OpenCV库的主要模块。如果您是PyImageSearch阅读器，则不需要安装此软件包。
opencv-contrib-python：包含主要模块与contrib模块。 # 一般推荐安装，因为它包含了所有的OpenCV功能。
opencv-python-headless：与opencv-python相同但没有GUI功能。适用于无界面系统。
opencv-contrib-python-headless：与opencv-contrib-python相同，但没有GUI功能。适用于无界面系统。

运行demo

# 官方示例
$ python projects/CenterNet2/demo/demo.py --config-file projects/CenterNet2/configs/CenterNet2_R50_1x.yaml --input path/to/image/ --opts MODEL.WEIGHTS models/CenterNet2_R50_1x.pth

# 运行代码
$ python /home/leng/Documents/nemo/CenterNet2-master/projects/CenterNet2/demo.py --config-file /home/leng/Documents/nemo/CenterNet2-master/projects/CenterNet2/configs/CenterNet2_R2-101-DCN-BiFPN_4x+4x_1560_ST.yaml --input /home/leng/Documents/nemo/temp_img/ --output /home/leng/Documents/nemo/temp_output/ --opts MODEL.WEIGHTS /home/leng/Downloads/CenterNet2_R2-101-DCN-BiFPN_4x+4x_1560_ST.pth

# 使用tensorborad查看loss等信息
$ tensorboard --logdir /home/leng/Documents/T42/detectron2/output/CenterNet2/Centernet2_3090

第一次使用时的报错，记录一下：

/home/leng/anaconda3/envs/detectron2/lib/python3.8/site-packages/torch/cuda/__init__.py:104: UserWarning: 
NVIDIA GeForce RTX 3090 with CUDA capability sm_86 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 compute_37.
If you want to use the NVIDIA GeForce RTX 3090 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

  warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))

AttributeError: module ‘distutils‘ has no attribute ‘version‘ 解决方案

1 问题 PyTorch版本1.10，安装完tensorboard后运行代码时出现错误：AttributeError: module 'distutils' has no attribute 'version'。

2 原因 setuptools版本问题。

3 解决方案

pip uninstall setuptools
pip install setuptools==59.5.0