Closed shivamsaboo17 closed 5 years ago
It seems plausible that an l1 loss could work for OOD segmentation. (We will likely put up a paper in a month or two proposing a dataset for the task of OOD segmentation.)
If possible, can you tell me if the labels for segmentation of OOD samples would be random (similar to classification task) or something else?
I would first try giving anomalies a k+1st class.
On Sun, Aug 25, 2019 at 11:38 PM Shivam Saboo notifications@github.com wrote:
If possible, can you tell me if the labels for segmentation of OOD samples would be random (similar to classification task) or something else?
— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/hendrycks/outlier-exposure/issues/6?email_source=notifications&email_token=ACZBITRD4PN7XW7LDBWE3M3QGN3ABA5CNFSM4IK7W6E2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5DNPRQ#issuecomment-524736454, or mute the thread https://github.com/notifications/unsubscribe-auth/ACZBITTQFYTI5LKBV6KXB23QGN3ABANCNFSM4IK7W6EQ .
I was going through the paper and in section 3 where the OE minimization objective is defined, there is no constraint on the type of loss that is being used. Hence I was curious if the methodology can be applied to tasks where we do pixel-level prediction (example image translation) and use L1 or L2 loss? Similar to classification loss, if we use uniform random ground truth images as the label for ood images, will it work or dealing with pixel-level predictions, require a completely different methodology?