https://github.com/datawhalechina/team-learning-cv/blob/master/AerialImageSegmentation/Task1%EF%BC%9A%E8%B5%9B%E9%A2%98%E7%90%86%E8%A7%A3.md

    https://tianchi.aliyun.com/competition/entrance/531872/information

    目标:
    image.png



    理论:
    1、理解评价指标 Dice coefficient
    image.png


    步骤:
    1、下载数据
    windows下,先wsl切换到到linux terminal,之后:
    mkdir data && cd data
    wget http://tianchi-competition.oss-cn-hangzhou.aliyuncs.com/531872/%E5%9C%B0%E8%A1%A8%E5%BB%BA%E7%AD%91%E7%89%A9%E8%AF%86%E5%88%AB/train_mask.csv.zip
    wget http://tianchi-competition.oss-cn-hangzhou.aliyuncs.com/531872/%E5%9C%B0%E8%A1%A8%E5%BB%BA%E7%AD%91%E7%89%A9%E8%AF%86%E5%88%AB/train.zip
    wget http://tianchi-competition.oss-cn-hangzhou.aliyuncs.com/531872/%E5%9C%B0%E8%A1%A8%E5%BB%BA%E7%AD%91%E7%89%A9%E8%AF%86%E5%88%AB/test_a_samplesubmit.csv
    wget http://tianchi-competition.oss-cn-hangzhou.aliyuncs.com/531872/%E5%9C%B0%E8%A1%A8%E5%BB%BA%E7%AD%91%E7%89%A9%E8%AF%86%E5%88%AB/test_a.zip
    unzip train_mask.csv.zip test_a_samplesubmit.csv test_a.zip
    2、读取数据:
    image.png

    3、测试提交结果

    安装环境:
    查看cuda版本:
    nvida控制面板里-》系统信息
    image.png

    conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch



    解决conda下载慢问题:
    conda config —add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
    conda config —add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
    conda config —set show_channel_urls yes

    来自 [https://blog.csdn.net/qq_42514225/article/details/107736711](https://blog.csdn.net/qq_42514225/article/details/107736711)


    几个问题:

    • 理解RLE编码过程,并完成赛题数据读取并可视化;
    • 回顾下pandas的一些操作:

    统计所有图片整图中没有任何建筑物像素占所有训练集图片的比例;
    统计所有图片中建筑物像素占所有像素的比例;
    计所有图片中建筑物区域平均区域大小;

    • FCN模型