Possible improvement course

[manerotoni]

Show mask data on the first exercise, cell classification, in addition of the bounding boxes

[manerotoni]

Add in between some task for the students. For instance plot an additional image for looking at the data

[manerotoni]

Resize and padding. A resize of images is probably not a good idea if a phenotype makes very obligated cells instead of round.
this could be a small idea of thinking about it for the students

[manerotoni]

May be a little confusing on how training loss and validation loss plot. One is running average per iteration the other is per epoch

[manerotoni]

The weight exercises should be more explained. Also the syntax on how to do
something like
w1 =1 w2 =1 torch.tensor([w1, w2])
The student should modify w1 and w2 in a meaningful way.

[manerotoni]

Exercise for the segmentation with self-trained model could be to adapt image size to match the network

[manerotoni]

Exercise of the U-Net receptive field could be to input a pixel and look how it propagates

[constantinpape]

Add more image analysis to exercise 3: also apply classification model and measure the serum intensities of infected and non-infected cells.

[manerotoni]

Make listing of files OS independent. For instance do not use ls but more os or glob