Models for Self-Supervised Downstream Detection Tasks

[chos1npc]

I would like to ask whether it is possible to train the model transferred from self-supervision through mmdet. Is the naming of the backbone consistent with most self-supervised models?

[hhaAndroid]

@chos1npc Self-supervised training is currently done using mmpretrain, and the resulting models can be directly used in MMDetection. You can perform self-supervised training using the mmpretrain library and then directly utilize the trained models in the MMDetection framework for tasks such as object detection.

https://github.com/open-mmlab/mmpretrain/blob/main/docs/en/user_guides/downstream.md

[chos1npc]

@hhaAndroid Is it possible to transfer a self-supervised model that wasn't trained using mmpretrain?

[hhaAndroid]

@chos1npc Of course, you can simply transfer some keys or directly copy the backbone code and register it to achieve that.