A simple implementation of LeNet5 for practice for the book "Pytorch Pocket Reference"
LeNet is abunch of convolution and linear layers with max pools.
- Lucidrains
- Agorians
pip install lenet5
import torch
from lenet5 import LeNet5
x = torch.randn(1, 3, 32, 32)
model = LeNet5()
result = model(x)
print(result)
print(result.shape)
print(result.dtype)The LeNet5 architecture is composed of:
=> 2 convolutional layers with varying inc hannels and kernel sizes.
=> Linear layers
=> 2x max pool2d layers applied on relu -> convolutional layers(x)
=> view, resizes according to -1 and the int of the element of the first dimension
=> 2 relus on the linear layers respectively
=> final linear projection
MIT
GradientBased Learning Applied to Document Recognition Yann LeCun Leon Bottou Yoshua Bengio and Patrick Hafner http://vision.stanford.edu/cs598_spring07/papers/Lecun98.pdf