UNet++

This implements training of UNet on the 2018 Data Science Bowl dataset, mainly modified from UNet++.

UNet Detail

For details, see UNet++.

Requirements

• Install PyTorch (pytorch.org)

• pip install -r requirements.txt

• 若遇到module 'cv2' has no attribute '_registerMatType' 报错，可按照以下步骤解决

  pip install -r requirements.txt
  pip uninstall opencv-python
  pip uninstall opencv-python-headless
  pip install opencv-python-headless==4.6.0.66
  pip install tqdm
  pip install pillow==6.2.2
  

• get dataset from data-science-bowl-2018.The file structure is the following:

── data-science-bowl-2018
    ├── stage1_train
    |   ├── 00ae65...
    │   │   ├── images
    │   │   │   └── 00ae65...
    │   │   └── masks
    │   │       └── 00ae65...            
    │   ├── ...
    |
    ...

• the data-science-bowl-2018 dataset need preprocess.

python3 preprocess_dsb2018.py  --data_dir=./data-science-bowl-2018
数据会保存在./data-science-bowl-2018文件夹下

Training

一、训练流程

单卡训练流程：

1.安装环境  
2.开始训练
    bash ./test/train_full_1p.sh  --data_path=数据集路径         # 精度训练
    eg: bash ./test/train_full_1p.sh --data_path=./inputs/dsb2018_96/
    bash ./test/train_performance_1p.sh  --data_path=数据集路径  # 性能训练

​
多卡训练流程

1.安装环境
2.开始训练
    bash ./test/train_full_8p.sh  --data_path=数据集路径         # 精度训练
    eg:bash ./test/train_full_8p.sh --data_path=./inputs/dsb2018_96/
    bash ./test/train_performance_8p.sh  --data_path=数据集路径  # 性能训练

单卡或多卡在线评估：

	bash  ./test/eval.sh   数据集路径    模型文件路径

	eg: bash ./test/eval.sh ./inputs/dsb2018_96/ ./models/dsb2018_96_NestedUNet_woDS/model_best.pth.tar

demo在线评估：

	python demo.py --data_path=./inputs/dsb2018_96

二、Docker容器训练

1.导入镜像二进制包docker import ubuntuarmpytorch.tar REPOSITORY:TAG, 比如:

docker import ubuntuarmpytorch.tar pytorch:b020

2.执行docker_start.sh后带三个参数：步骤1生成的REPOSITORY:TAG；数据集路径；模型执行路径；比如：

./docker_start.sh pytorch:b020 /train/peta /home/DeepMar

3.执行步骤一训练流程（环境安装除外）

三、测试结果 训练日志路径：网络脚本test下output文件夹内。例如： test/output/devie_id/train_${device_id}.log # 训练脚本原生日志 test/output/devie_id/UNet++_bs1024_8p_perf.log # 8p性能训练结果日志 test/output/devie_id/UNet++_bs1024_8p_acc.log # 8p精度训练结果日志

训练模型：训练生成的模型默认会写入到和test文件同一目录下。当训练正常结束时，checkpoint.pth.tar为最终结果。

UNet++ training result