search

Luodian / RelateAnything

Relate Anything Model is capable of taking an image as input and utilizing SAM to identify the corresponding mask within the image.
Apache License 2.0
437 stars 20 forks source link

About self-distillation training strategy #14

Closed Vincent-luo closed 5 months ago

Vincent-luo commented 5 months ago

Thanks for your excellent work! And I have a question about the training strategy. In the report you mentioned, it uses self-distillation with soft labels. I wonder if you used this to train the RAM model and how much improvement will this technique bring?

franciszzj commented 5 months ago

Thank you for your interest in our project. We did not use self-distillation with soft labels. We only used the labels annotated in the PSG dataset to train the Relate Anything model.