search

tstandley / taskgrouping

Code for Which Tasks Should Be Learned Together in Multi-task Learning?
Other
93 stars 14 forks source link

Visualising the error maps? #4

Open deblinaml opened 3 years ago

deblinaml commented 3 years ago

Hi could you please share the script using which you visualised the error maps in the paper? Thanks in advance!

tstandley commented 2 years ago

Hi. Sorry for the late reply. I lost access to github for a while around when you asked, and when I got it back I forgot to give you an answer.

I have not released that code and have no plans to. It's extremely messy and I'd have to dig around in archives to even find it.

The error maps for the tasks other than SemSeg are fairly straightforward. I just plot the absolute difference between ground truth and prediction and use a color map.

For SemSeg, I think I actually take the maximum class (max between ground truth and prediction) whenever they don't agree.

Hope this helps!