pytorch入门（1）

Tensorboard--SummaryWriter:

记录训练学习率和损失函数变化

使用方法：

from torch.utils.tensorboard import SummaryWriter

writer=SummaryWriter("logs")#创建实例
...
writer.close()

summarywriter参数：

log_dir:保存位置，文件夹名称

comment:备注

样例：

#create a summary writer with automatically generated folder name.
writer = SummaryWriter()
# folder location: runs/May04_22-14-54_s-MacBook-Pro.local/

# create a summary writer using the specified folder name.
writer = SummaryWriter("my_experiment")
# folder location: my_experiment

# create a summary writer with comment appended.
writer = SummaryWriter(comment="LR_0.1_BATCH_16")
# folder location: runs/May04_22-14-54_s-MacBook-Pro.localLR_0.1_BATCH_16/
# writer.add_image()

add_scalar()参数：

tag:标题

scalar_value:函数值

global_step:自变量

for i in range(100):
    writer.add_scalar('y=x',2*i,i)#添加标量数据到summary

add_image()参数：

tag：标题

img_tensor:图片数据类型，torch.Tensor, numpy.ndarray, or string/blobname

global_step：全局步长

dataformats:默认图片格式为（3，H，W）（输入为tensor类型），可设置CHW，HWC（numpy.ndarray），HW...

from torch.utils.tensorboard import SummaryWriter
import numpy as np
from PIL import Image
img_PIL=Image.open('hymenoptera_data/hymenoptera_data/train/ants/0013035.jpg')#用相对路径打开image
img_array=np.array(img_PIL)#PIL不符合格式要求
print(type(img_array))#<class 'numpy.ndarray'>
print(img_array.shape)#(512, 768, 3)
writer=SummaryWriter("logs")#创建实例
writer.add_image('test',img_array,1,dataformats='HWC')
writer.close()

pycharm 终端打开logs文件：

 当改变tag名称时，生成新的logs文件

Transform:

ToTensor:

将PIL类型、numpy.ndarray类型图片转换为tensor类型。

image读取图片，类型为PIL 

用python如何使用：

from torchvision import transforms
from PIL import Image
img=Image.open('hymenoptera_data/hymenoptera_data/train/ants/0013035.jpg')#相对地址
tensor_tran=transforms.ToTensor()#创建实例
tensor_img=tensor_tran(img)
print(tensor_img)

opencv读取图片，类型为numpy：

import cv2
cv_img=cv2.imread('hymenoptera_data/hymenoptera_data/train/ants/0013035.jpg')
print(type(cv_img))

<class 'numpy.ndarray'>

用image读取图片，并生成logs文件，完整流程：

from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms
from PIL import Image
img=Image.open('hymenoptera_data/hymenoptera_data/train/ants/0013035.jpg')#相对地址 PIL类型
tensor_tran=transforms.ToTensor()#创建实例
tensor_img=tensor_tran(img)#将PIL转换为tensor
writer=SummaryWriter('logs')
writer.add_image('tensor_img',tensor_img)
writer.close()

Normalize：

#Normalize归一化
norm_trans=transforms.Normalize([1,2,3],[3,2,1])#均值、标准差
img_norm=norm_trans(img_tensor)
print(img_tensor[0][0][0])
print(img_norm[0][0][0])
writer.add_image('Normalize',img_norm,4)

normalize效果：

(input-0.5)/0.2=2*input-1
 input[0,1]
 result[-1,1]

Resize按比例缩放：

print(img.size)
trans_resize=transforms.Resize((512,512))#图片为PIL型
#img_resize->resize->img_resize PIL
img_resize=trans_resize(img)
img_resize=trans_tensor(img_resize)
writer.add_image('resize',img_resize,0)
print(img_resize)

Compose：

trans_resize_2=transforms.Resize(512)
trans_compose=transforms.Compose([trans_resize_2,trans_tensor])#PIL-PIL PIL-tensor 输出与下一个输入匹配
#Compose([transforms参数1],transforms参数2)
img_compose=trans_compose(img)
writer.add_image('compose',img_compose,1)

RandomCrop随机裁剪：

rans_random=transforms.RandomCrop(300)
trans_compose_2=transforms.Compose([trans_random,trans_tensor])
for i in range(10):
    img_crop=trans_compose_2(img)
    writer.add_image('Randomcrop',img_crop,i)

【PyTorch深度学习快速入门教程（绝对通俗易懂！）【小土堆】】https://www.bilibili.com/video/BV1hE411t7RN?p=9&vd_source=60ec93acfa1fc770b0d4a815c06d6583