search

ethz-asl / background_foreground_segmentation

12 stars 2 forks source link

Feature/ignorant loss #6

Closed renezurbruegg closed 3 years ago

renezurbruegg commented 3 years ago

Created custom loss function to ignore classes with class label 1 (=Not labeled)

I also created a utils package and moved the utils.py function there. Might cause problems in other training code.

renezurbruegg commented 3 years ago

I also created a pytest test. Would be nice to run pytest on build server, not sure if this is happening right now

renezurbruegg commented 3 years ago

I moved the changes from the image preprocessing branch to this pull request here. The build seems to work now.

hermannsblum commented 3 years ago

I also created a pytest test. Would be nice to run pytest on build server, not sure if this is happening right now

I guess it can be just run with pytest . to execute all tests? or does this work differently? I can then add this to the CI settings.

renezurbruegg commented 3 years ago

I also created a pytest test. Would be nice to run pytest on build server, not sure if this is happening right now

I guess it can be just run with pytest . to execute all tests? or does this work differently? I can then add this to the CI settings.

Yes, or using pytest tests since pytest . also runs the NYU tests which fail on my machine

renezurbruegg commented 3 years ago

Did the same thing for the accuracy + added a balanced accuracy metric.

Did a pull request, however creating a new pull requests based on main can not be built somehow.

hermannsblum commented 3 years ago

@renezurbruegg The build works now and is also running the tests (please take note that I moved the location of test scripts closer to the code). However, there is a conflict with dataLoader which was removed on this branch. What is the reason?

renezurbruegg commented 3 years ago

@renezurbruegg The build works now and is also running the tests (please take note that I moved the location of test scripts closer to the code). However, there is a conflict with dataLoader which was removed on this branch. What is the reason?

Thanks!

I did not remove it on purpose. I created this branch from the master branch when there was no dataloader on it. Since the losses do not use the loader I did not push it to this branch in order to not have more code to review.

I'll merge the master into this branch asap

renezurbruegg commented 3 years ago

Was using my girlfriends computer to merge 0c30b2d39b0801a9d5a6e3d32ee7550b757c7534 if you are wondering about the username :D

renezurbruegg commented 3 years ago

run linter