chapter7

torchvision's architecture

The main method of transforms.Compose()

• Scale
• RandomHorizontalFlip
• RandomCrop

Understanding of residual network:

• Why is it proposed
  The previous assumptions were based on nested structures, but the following non-nested structures will appear
  

As the number of layers of the neural network increases, although the range of functional relationships that can be fitted will increase (that is, the area covered by the last layer of neural network), it is very likely that it will become more and more difficult to fit.The closest function to f*.

The residual module forces the structure to be transformed into a nested structure

torch.max
This definition comes from pytorch

def max(self: Tensor, dim: _int, keepdim: _bool=False, *, out: Optional[Tensor]=None) -> Tuple[Tensor, Tensor]: …

Tensor: Given a tensor
keepdim: dimension (0, 1: row, column)
For a tensor with the maximum value within the given dimension