Summary (summer)

Programming experience:

• Python placed "//" used is rounded down
• Program given: "iteration over a 0-d tensor", probably because the function returns the value of a parameter of the function call in the process but it takes two parameters. 
• cv2.imread to read the picture format is BGR, PIL to read the picture is the RGB format, two ways to read the picture in the show when there is a difference
  • How to convert RGB to BGR image format image format it?
    • method one

      • b,g,r = cv.split(img)#拆分通道
        img_1 = cv.merge([r,g,b])#合并通道

    • Method Two 

      • img_2 = img[:,:,::-1]

• Pytorch in ImageFolder data read is a better way
• Pytorch detach the difference in the detach_ (codes can be appreciated by the GAN)
• DCGAN when using Adam updating optimizer, arranged to be noted beta1 = 0.5

Theoretical summary:

• 3 ways to enhance the receptive field:
  • Stacking a plurality of layers convolution
  • The introduction of pooling layer, but can cause loss of information
  • Convolution using empty