-
Notifications
You must be signed in to change notification settings - Fork 284
Open
Description
DFINE-N was pretrained with COCO2017
1: person
2: bicycle
3: car
4: motorcycle
...
I want to do fine-tuninig with custom coco data.
that custom coco data has partial coco classes, for example,
1: person
4: motorcycle
...
But in my fine-tuining test, DFINE-N requires continuous class index (starting from 0) like
0: person
1: motorcycle
...
So, if I use this dataset (0: person, 1: motorcycle, ...) for fine-tuning DFINE-N, there is class index mismatch.
How to resolve this class index mismatch?
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels