search

mdbloice / Augmentor

Image augmentation library in Python for machine learning.
https://augmentor.readthedocs.io/en/stable
MIT License
5.08k stars 866 forks source link

Random Ereasing areas too big #242

Closed MasIgor closed 2 years ago

MasIgor commented 2 years ago

Hello!

when using random erasing i cant go below 0.11 size. 11% of the size is too big for me, i would need a tenth of this. can this be adapted somehow?

thank you!

best regards

Igor

mdbloice commented 2 years ago

Hi @tanzerlana yes that's a current limitation, patches need to be >0.1 and <1.0...

How small would you need your patches to be?

MasIgor commented 2 years ago

Hello! thanks for your kind answer..

for my application I would need them to be between 0.01 and 0.1 as I am adding multiple labels it fills up too much and does not simulate our real world scenario

Thank you!

mdbloice commented 2 years ago

Hi again @tanzerlana - if you update Augmentor using pip you should be able to now use values between 0.01 and 1.

Let me know if it's now working as you'd expect!

MasIgor commented 2 years ago

Greaaaat! Thank you! Will test it asap

Von: Marcus D. Bloice @.> Gesendet: Mittwoch, 27. April 2022 11:39 An: mdbloice/Augmentor @.> Cc: Igor Masin - Tanzer Maschinenbau GmbH @.>; Mention @.> Betreff: Re: [mdbloice/Augmentor] Random Ereasing areas too big (Issue #242)

Hi again @tanzerlanahttps://github.com/tanzerlana - if you update Augmentor using pip you should be able to now use values between 0.01 and 1.

Let me know if it's now working as you'd expect!

— Reply to this email directly, view it on GitHubhttps://github.com/mdbloice/Augmentor/issues/242#issuecomment-1110789319, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQGO4ZYYGB4C5ULDF5JEG3TVHEDMLANCNFSM5UOFVBMA. You are receiving this because you were mentioned.Message ID: @.**@.>>

MasIgor commented 2 years ago

@mdbloice i did not forget you, i am at home in quarantene and will be back on monday to test it.