-
Notifications
You must be signed in to change notification settings - Fork 435
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Added Self-SL adapter for classification to handle case when only images for train data provided + Documentation update #2219
Added Self-SL adapter for classification to handle case when only images for train data provided + Documentation update #2219
Conversation
The idea is to add fake label to "image dir" dataset format for classification in case of Self-SL when no annotations are provided. This will let us to construct OTX dataset and proceed with Byol unsupervised training providing just folder with any images |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Overall, LGTM. BTW, how about setting the default training_type as incremental
?
And triggering auto mode by inserting --train-type auto
?
I know this is just my preference, so this is a minor comment.
In my opinion, middle~high level developer assumes the supervised
manner is default.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, but I have a question.
Do users should set only folder with images to use self-sl?
What about if users want self-sl w/ class folders?
Sure, users can run Self-SL with image folder dataset as before. However, in that case we need to add |
Setting train_type as None by default is solution how to distinguish between actual incremental passed by user in CLI and just default value. As we discussed on the meeting it is the most simple and working solution for that. Now, there is no overlap with corner cases and using auto train type detection is safe. I documented also one corner case, when we pass image folder format to SelfSL task instead of images only. In that case we have overlap with auto split for incremental. I also added this note in documentation both in explanation and tutorial sections. |
@kprokofi Thanks for answering. Let's take a look at test issues. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the explanation, before merging, we need to resolve test failures.
03e2fcb
to
b2b2122
Compare
Summary
How to test
Checklist
License
Feel free to contact the maintainers if that's a concern.