CNN Architecture design

Hi , If I am designing a cnn architecture from scratch , How should I decide how many layers to use. Can I anticipate it in advance depending on the dataset for ex - imagenet or MNIST or I will have to do trial and error .

Experimentation is the most common design method. Add only as much complexity as is needed to get “good enough” results.


As @TMosh said, experimentation is key to find a good model. Try simple ones first until you get the performance you are looking for. You can always check the most popular models, like LeNet, VGG, ResNet and EfficientNet families, etc

1 Like