[AI Combat] Small target detection model SSPNet--training environment built from scratch

Introduction to SSPNet

SSPNet: Scale Selection Pyramid Network for Tiny Person Detection from UAV Images

• Official link
  https://github.com/MingboHong/SSPNet
• arxiv
  https://arxiv.org/abs/2107.01548

Environment build

The cuda of my machine is 10.2. According to your own cuda version, go to https://hub.docker.com/ to pull the corresponding image, and you
must pull the version image with devel

• Pull the gpu image

docker pull aegis1/cuda10.2-cudnn8-devel-ubuntu18.04:pcl

• create container

nvidia-docker run -it -d \
    --name sspnet \
    -v /bee/abc/test_model/:/notebooks \
    -e TZ='Asia/Shanghai' \
    --shm-size 16G \
    -d aegis1/cuda10.2-cudnn8-devel-ubuntu18.04:pcl

• into the container

 docker exec -it sspnet env LANG=C.UTF-8 /bin/bash

install dependencies

• Install Anaconda

curl -O https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
sh Miniconda3-latest-Linux-x86_64.sh
ln -s /root/miniconda3/bin/conda /usr/bin/conda

• Create a py3.8 environment

conda create --name SSPNet python=3.8 -y
conda init

Exit the terminal
and enter the terminal again,
the following content will appear:

(base) root@1212:/#

• Enter our environment SSPNet

conda deactivate
conda activate SSPNet

As follows;

(SSPNet) root@1212:

• Install torch
  my cuda is 10.02

conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=10.2 -c pytorch

Other cuda versions can be installed according to
https://pytorch.org/get-started/previous-versions/

• install mim

pip install openmim

• Install mmcv-full

mim install mmcv-full==1.3.18 

• Install cv2 dependencies

apt update
apt install libgl1-mesa-glx  -y

• Install SSPNet

git clone https://github.com/MingboHong/SSPNet.git
cd SSPNet
pip install -r requirements.txt
python setup.py develop

【】Follow the above process to install step by step, you can run the model normally

reference

• https://github.com/MingboHong/SSPNet
• https://mmcv.readthedocs.io/en/latest/get_started/installation.html#install-mmcv
• https://mmdetection.readthedocs.io/zh_CN/latest/get_started.html
• https://github.com/open-mmlab/mim