search

slds-lmu / lecture_i2dl

Introduction to Deep Leaning
Creative Commons Attribution 4.0 International
2 stars 0 forks source link

Misleading figure in slides-cnn-cnn-components #27

Open AsjaFischer opened 1 year ago

AsjaFischer commented 1 year ago

The CNN figure on slide 11 is misleading. The filter weights of for all channels are the same, which is not the case in practice!

davidruegamer commented 1 year ago

Thanks, that's also correct. Since we are anyway missing some important basic information about how CNNs work w.r.t. different channels, I would revise this part anyway including some new materials.

AsjaFischer commented 1 year ago

That sounds good! We also have a slide "working with images" in the 2-D conv slides. I would suggest to remove it there and copy some of the content after the imput-channel-slides of this section followed by an image like this https://i.stack.imgur.com/3m7mW.png taken from this http://cs231n.github.io/convolutional-networks/.

Then we could remove the last slides from this lecture. What do you think ?