ImageNet Pretrained Weight in OWOD

[childult-programmer]

Hello. Firstly, kudos to you for your research on the OWOD task. I have a question about utilizing ImageNet pre-trained weights in OWOD.

I thought that since the premise of OWOD is that the initial detection framework should not know about Unknown class objects, it would be more correct to have all frameworks (even the feature extraction phase) randomly initialized without using Imagenet pretrained weights.

If I'm wrong in my understanding of the OWOD task, or if you can explain the premise of the OWOD task, I'd appreciate it.