1. 安装

pytorch conda环境

pip install tensorboard
pip install opencv-python

2. add_scalar 查看函数图形

常用来查看 train val loss等函数图形

from torch.utils.tensorboard import SummaryWriter

writer = SummaryWriter("logs")

for x in range(100):
    y = x
    writer.add_scalar("title: y=x", y, x)

writer.close()

执行代码查看结果

3. 查看结果

tensorboard --logdir=logs --port=6007

如果logs目录不清空，多次执行的函数图形都会显示，如果title名相同，两次执行的结果还会显示在同一张图上，并会对两次结果做拟合，如果只想显示当前函数的图像，则需要将logs目录清空，重现执行上面的shell命令或者刷新

4. add_image() 查看训练步骤中间结果的图片

opencv打开图片，数据是numpy数组
PIL打开图片，数据是PIL图片类型
也可以用numpy.array 处理图片数据

https://blog.csdn.net/qq_42751978/article/details/130720186

from torch.utils.tensorboard import SummaryWriter
from cv2 import imread
from PIL import Image
import numpy as np
writer = SummaryWriter("logs")
img_path = "hymenoptera_data/train/bees/21399619_3e61e5bb6f.jpg"
img_cv2 = imread(img_path)   # ndarray

img_PIL = Image.open(img_path)
img_array = np.array(img_PIL)

writer.add_image("train", img_cv2, 2, dataformats='HWC')
writer.add_image("train", img_array, 3, dataformats='HWC')
for x in range(100):
    y = 3*x
    writer.add_scalar("title: y=x", y, x)

writer.close()