# DenseNet_Train_Model

**Repository Path**: xbw12138/DenseNet_Train_Model

## Basic Information

- **Project Name**: DenseNet_Train_Model
- **Description**: DenseNet Train Model
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2019-12-19
- **Last Updated**: 2024-03-22

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# ICIAR2018_BACH_Challenge

## 1染色归一化

### 1.1配置环境
```
conda create -n color
conda activate color

pip install numpy
pip install opencv-python
```

### 1.2文件目录
在数据集中挑选一张图像做为source.tif
在Photos目录下新建Normalization文件夹存放染色归一化后的数据集

### 1.3执行

```
chmod +x batch.sh
./batch.sh
```

-------

## 2图像分割

### 2.1配置环境
```
conda create -n color
conda activate color

pip install numpy
pip install opencv-python
```

### 2.2文件目录
在Photos目录下新建Patches文件夹存放分割切片后的数据集，分割大小512*512 步长256

### 2.3执行

```
python split.py -i Photos/Normalization/Benign -o Photos/Patches/Benign
python split.py -i Photos/Normalization/InSitu -o Photos/Patches/InSitu
python split.py -i Photos/Normalization/Invasive -o Photos/Patches/Invasive
python split.py -i Photos/Normalization/Normal -o Photos/Patches/Normal
```

-------

## 3数据增强
### 3.1配置环境
```
conda create -n color
conda activate color

pip install numpy
pip install opencv-python
```
### 3.2文件目录
在Photos目录下新建Augmentation文件夹存放旋转，翻转后的数据集，1原图，2逆时针90度，3逆时针180度，4逆时针270度，5垂直翻转，6水平翻转

### 3.3执行
```
python augment.py -i Photos/Patches/Benign -o Photos/Augmentation/Benign
python augment.py -i Photos/Patches/InSitu -o Photos/Augmentation/InSitu
python augment.py -i Photos/Patches/Invasive -o Photos/Augmentation/Invasive
python augment.py -i Photos/Patches/Normal -o Photos/Augmentation/Normal
```

-------

## 4合并图像分割与数据增强

### 4.1说明
需要在染色归一化后划分数据集，60:20:20，训练集/验证集/测试集，所以归一化后最好一步到位。

### 4.2配置环境
```
conda create -n color
conda activate color

pip install numpy
pip install opencv-python
```
### 4.3文件目录
在Photos目录下新建Augmentation文件夹存放直接从染色归一化后的图像旋转，翻转后的数据集，1原图，2逆时针90度，3逆时针180度，4逆时针270度，5垂直翻转，6水平翻转

### 4.4执行
```
python split_augment.py -i Photos/Normalization/train/Benign -o Photos/Augmentation/Benign
python split_augment.py -i Photos/Normalization/train/InSitu -o Photos/Augmentation/InSitu
python split_augment.py -i Photos/Normalization/train/Invasive -o Photos/Augmentation/Invasive
python split_augment.py -i Photos/Normalization/train/Normal -o Photos/Augmentation/Normal
```

-------

## 5采用DenseNet网络训练模型

### 5.1配置环境

```
conda create -n color
conda activate color
conda install pytorch
pip install torchvision
```

### 5.2参数说明
Densenet for ICIAR2018

positional arguments:
  action                输入train or test
optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET, -d DATASET 数据集路径
  --depth DEPTH, -t DEPTH DenseNet深度，默认40
  --growth_rate GROWTH_RATE, -g GROWTH_RATE DenseNet增长率，默认12
  --num_classes NUM_CLASSES, -n NUM_CLASSES 几分类，默认4
  --epoch EPOCH, -e EPOCH 训练轮数，默认200
  --batch_size BATCH_SIZE, -b BATCH_SIZE 数据批数，默认20，即一轮训练20条数据
  --ckpt CKPT, -c CKPT  训练模型保存文件,以及测试集需要模型，默认model.pth

### 5.3执行

```
python train.py -d Photos/Augmentation/ train
```

-------

## 6数据随机增强随机分割
### 6.1说明
1. 使用两种参数，随机分割为512 * 512或 1024 * 1024
2. 每种图像生成12张图像

### 6.2配置环境

```
conda create -n color
conda activate color
conda install pytorch
pip install torchvision
```

### 6.3执行
1. 分割1024大小

```
python random_augment_split.py -i Photos/Normalization/train/Benign -o Photos/Augmentation_Random_1024_12/Benign -s 1024 -n 12

python random_augment_split.py -i Photos/Normalization/train/InSitu -o Photos/Augmentation_Random_1024_12/InSitu -s 1024 -n 12

python random_augment_split.py -i Photos/Normalization/train/Invasive -o Photos/Augmentation_Random_1024_12/Invasive -s 1024 -n 12

python random_augment_split.py -i Photos/Normalization/train/Normal -o Photos/Augmentation_Random_1024_12/Normal -s 1024 -n 12
```

2. 分割512大小

```
python random_augment_split.py -i Photos/Normalization/train/Benign -o Photos/Augmentation_Random_512_12/Benign -s 512 -n 12

python random_augment_split.py -i Photos/Normalization/train/InSitu -o Photos/Augmentation_Random_512_12/InSitu -s 512 -n 12

python random_augment_split.py -i Photos/Normalization/train/Invasive -o Photos/Augmentation_Random_512_12/Invasive -s 512 -n 12

python random_augment_split.py -i Photos/Normalization/train/Normal -o Photos/Augmentation_Random_512_12/Normal -s 512 -n 12
```

-------


## 7数据随机增强随机分割（Plus）
### 7.1说明
size, num控制分割图像的大小与随机增强的个数

### 7.2配置环境

```
conda create -n color
conda activate color
conda install pytorch
pip install torchvision
```

### 7.3执行
1. 分割224大小，240张

```
python random_augment_split.py -i Photos/Normalization/train -s 224 -n 240
```
-------