Pretrained Model for AVA 32x2?

I have been experimenting with SlowFast for AVA-type tasks using the https://github.com/facebookresearch/SlowFast/blob/master/configs/AVA/SLOWFAST_32x2_R50_SHORT.yaml config. I wish to use the pretrained AVA models from the zoo and do some fine tuning, but it seems that the released weights are for different model configs (I get `Network weights not loaded` for every layer, due to size mismatching).

My questions:

- Are the maintainers able to release a model that aligns with the https://github.com/facebookresearch/SlowFast/blob/master/configs/AVA/SLOWFAST_32x2_R50_SHORT.yaml config?

- Can the maintainers document which AVA configs (from https://github.com/facebookresearch/SlowFast/blob/master/configs/AVA) the released AVA models (https://github.com/facebookresearch/SlowFast/blob/master/MODEL_ZOO.md#ava) were trained using and therefore can be loaded into?

- Is there documentation on the differences between the different configs found under https://github.com/facebookresearch/SlowFast/blob/master/configs/AVA ? I just picked the https://github.com/facebookresearch/SlowFast/blob/master/configs/AVA/SLOWFAST_32x2_R50_SHORT.yaml one arbitrarily :) Some configuration differences are mentioned in section 5.1 of https://openaccess.thecvf.com/content_ICCV_2019/papers/Feichtenhofer_SlowFast_Networks_for_Video_Recognition_ICCV_2019_paper.pdf but some comments at the top of each yaml file explaining the model configured in the file would be appreciated