[Docs] Custom Models?

[SenmiaoORZ]

The doc issue

The Custom models https://mmaction2.readthedocs.io/en/latest/advanced_guides/customize_models.html still under developing. I want to run a ava finetune task with a r3d_18 model(from torchvision).

MMaction2 has implementation for A Closer Look at Spatiotemporal Convolutions for Action Recognition. However, I think that is only for R2+1d.

I want to run this pretrained model on AVA fintuned tasks. However, I don't know how to define my custom models. Or will you support that model soon?

Suggest a potential alternative/fix

No response

[valentin-fngr]

Hey, I posted an answer that might help you here : https://github.com/open-mmlab/mmaction2/issues/2805#issuecomment-2031506177