In paper, there are 80 classes, and divided to 4 tasks. At each task, un-introduced instances would be labeled as 'unknown' (for ex: task 1: label index: 20-79 would be labeled as 'unknown'), and the following task would add instances of these 'unknown' classes to continue training.
My question is: I do not pre-labeled any instances of new label so my dataset does not have any annotations with new label category. Can I use method proposed in paper to label instances of new label as 'unknown' .
If I can, what do I have to do or which files I need to change?
Thanks so much for your support!
In paper, there are 80 classes, and divided to 4 tasks. At each task, un-introduced instances would be labeled as 'unknown' (for ex: task 1: label index: 20-79 would be labeled as 'unknown'), and the following task would add instances of these 'unknown' classes to continue training.
My question is: I do not pre-labeled any instances of new label so my dataset does not have any annotations with new label category. Can I use method proposed in paper to label instances of new label as 'unknown' .
If I can, what do I have to do or which files I need to change?
Thanks so much for your support!