Lesson 7
- fist half in rnn
batch normalization
semantic segmentation
Computer vision - starts at 1:02
Cifar 10 - small data set of images with images of small sizes
stats - mean and standard deviation of channel (we need this no normalise data)
tfms - transform data
line 9 - creating a customer learner just a fully connected model
fully convelutional network
create a CNN- convolution
stride 2 conv. we move the kernel by two so we end up halving the resolution
adaptive max pool- last layer we do a maxpool just specify what size we want the output image second last we make a 1x1 max pool
then put it in a linear layer with output to amount of classes
padding- so we can get the feature at the edges of the image
pick up lesson at 1:56