Storing functions commonly used in training pytorch models
# 验证集分割 数据增强 记录损失曲线 AdaBound优化器 动态显示训练过程(类似caffe) 自动设置线程数 图片异常跳过
Solution(model, # the model need training 训练用的模型
criterion, # loss function 损失函数
optimizer, # default: AdaBound 优化器：默认为AdaBound
image_path, # all image folder 所有图片的文件夹
scheduler=None, # 调度器
epochs=100, # 训练轮次
batch_size=64, # 训练批次
show_interval=20, # show loss and accuracy per training show_interval batches 每训练show_interval个批次显示一次损失与准确率
valid_interval=100, # test the model per valid_interval batches 每训练valid_interval个批次测试一次模型
record_loss=True, # Whether to use tensorboardX to record the training process 是否使用tensorboardX记录训练过程
image_size=224, # image input size 图片输入尺寸
thread_size=8, # default: number of cpu 线程数：默认为cpu核心数
seed=66, # random state 随机数种子
ten_crops=False, # Whether to use TenCrops data augmentation 是否使用TenCrops增强训练数据
test_ratio=0.25, # default: 1/4 images in image folder will be used to test 图片文件夹中用于验证集的比率
)
# images should be placed in folders like:
# --root
# ----root\dogs
# ----root\dogs\image1.png
# ----root\dogs\image2.png
# ----root\cats
# ----root\cats\image1.png
# ----root\cats\image2.png
# path: the root of the image folder
# Split the entire picture folder into a training set and a validation set, with a directory structure lik: root directory {the directory for each category {the corresponding pictures}}
# 分割整个图片文件夹为训练集和验证集，目录结构为根目录{每个类别的目录{对应类别的图片}}
ImageFolderSplitter(path, # 所有图片文件夹
train_size=0.8, # 图片文件夹中用于训练集的比率
seed=66,
drop_type=('txt', 'csv'), # skip file with these suffix 遍历图片目录时需要跳过的文件类型
)