Article on 'bag of tricks' for NN training...cutout occlusion

Just read this article and thought it might be of interest:

FastAI already uses many of the tricks noted, but one item they talked about was cutout occlusion (covering up random portions of the image as a form of data augmentation). Maybe this is something we should try and add in as a transform for FastAI?


+1 for adding cutout occlusion

could be super useful for Deep Learning with Audio

1 Like