Pytorch搭建工程参考资料

Pytorch搭建工程参考资料

pytorch搭建完整的工程一般包括：数据加载、网络模型、训练、测试、损失函数等。以下列出我初次学习搭建工程所参考的资料。

数据加载

使用Dataset https://blog.csdn.net/sinat_42239797/article/details/90641659?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-2.nonecase&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-2.nonecase
使用Dataset或者torchvision https://blog.csdn.net/qq_39938666/article/details/88429064
Pytorch数据读取(Dataset, DataLoader, DataLoaderIter) https://zhuanlan.zhihu.com/p/30934236
Pytorch对两张图片进行相同的数据增广操作 https://blog.csdn.net/happyeveryday62/article/details/104350332

读取Tiff格式图像

Python 读取图片（非opencv） https://www.cnblogs.com/yinxiangnan-charles/p/5928689.html
Python 获取图片像素 https://blog.csdn.net/search7/article/details/75528537
Pytorch之浅入torchvision.transforms.ToTensor与ToPILImage https://blog.csdn.net/qq_37385726/article/details/81771980
Python：读和写TIFF 16位，三通道，彩色图像 来>自 https://cloud.tencent.com/developer/ask/52680

损失函数

pytorch自定义损失函数custom loss function https://blog.csdn.net/dss_dssssd/article/details/84103834

加载模型

Pytorch迁移学习加载部分预训练权重 https://blog.csdn.net/xzy5210123/article/details/88598436
pytorch加载模型和初始化权重 https://www.jianshu.com/p/7a7d45b8e0ee

训练

PyTorch学习之六个学习率调整策略 https://blog.csdn.net/shanglianlm/article/details/85143614?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-3.nonecase&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-3.nonecase
torch.optim.lr_scheduler：调整学习率 https://blog.csdn.net/qyhaill/article/details/103043637
Pytorch中常用的四种优化器SGD、Momentum、RMSProp、Adam。 https://cloud.tencent.com/developer/article/1491393