Skip to content

Latest commit

 

History

History
98 lines (75 loc) · 1.57 KB

README_CN.md

File metadata and controls

98 lines (75 loc) · 1.57 KB
Raw

遥感图像场景分类

English Version

简介

RSCUP: 遥感图像场景分类

仓库路径应该组织成如下结构:

sense_classification/
    |->examples
    |->models
    |->prepare_data
    |->data
    |   |->rssrai_sense_cls
    |   |   |->train
    |   |   |->val
    |   |   |->test
    |   |->tf_records
    |   |->train_list
    |->ckpt
    |->tools

环境依赖

  1. tensorflow-gpu==1.12.0 (I only test on tensorflow 1.12.0)
  2. python==3.4.3
  3. numpy
  4. easydict
  5. opencv==3.4.1
  6. 有些包可能没列出来,根据错误提示安装

安装, 准备数据, 训练, 验证, 生成提交文件

安装

  1. 下载代码
git clone https://github.com/vicwer/sense_classification.git

准备数据

data目录结构:

data/
    |->rssrai_sense_cls
    |   |->train
    |   |->val
    |   |->test
    |   |->ClsName2id.txt
    |->train_list/train.txt
    |->tf_records

  1. 下载数据集并解压: train.zip, val.zip, test.zip, ClsName2id.txt

  2. 生成 tf_records:

cd tools
python3 img_encode.py

训练

${sense_classification_ROOT}目录提供了config.py, 可设置超参数

例如

cd ${sense_classification_ROOT}
vim config.py
cfg.train.num_gpus = {your gpu nums}
etc.

cd ${sense_classification_ROOT}/examples/
python3 multi_gpus_train.py

验证

cd ${sense_classification_ROOT}/examples/
python3 accuracy.py

生成提交文件

cd ${sense_classification_ROOT}/examples/
python3 submit.py

结果:

验证集: 0.908+ 测试集：0.90509