特征金字塔Transformer代码实现

1、源码地址
https://github.com/ZHANGDONG-NJUST/FPT
2、环境安装
本人是在服务器上安装的，版本Ubuntu 18.04
conda创建虚拟环境
首先将源改为清华源

conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/msys2/
conda config --set show_channel_urls yes

创建环境

conda create -n fpt python=3.6

激活环境

source activate fpt

【删除环境】这一步以备不时之需

conda remove -n fpt --all

安装pytorch、torchvision
pytorch=0.4.0
torchvision>=0.2.0
python 3.6.11环境下

conda install pytorch==0.4.0 torchvision==0.2.1 cuda100 -c pytorch

安装其余包
pip install cython
pip install matplotlib
pip install numpy
pip install scipy
pip install opencv-python
pip install pyyaml
pip install packaging
pip install dropblock==0.1.0
pip install pycocotools
pip install tensorboardX — for logging the losses in Tensorboard

Question：
gcc: error: /home// /ns-vqa/scene_parse/mask_rcnn/lib/model/nms/src/nms_cuda.c: No such file or directory
Solve:
https://github.com/kexinyi/ns-vqa/issues/1
https://github.com/jwyang/faster-rcnn.pytorch/blob/master/lib/model/nms/src/nms_cuda.c
缺少这个文件就下载下来
Question：
CompileError: command ‘gcc’ failed with exit status 1
Solve:
http://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
https://www.bbsmax.com/A/gGdX6NnQJ4/
https://github.com/jwyang/faster-rcnn.pytorch/issues/235