Is 2D RoI at a frame extended into a 3D RoI for AVA action detection task

The paper says
> We extract region-of-interest (RoI) features [6] at the last feature map of res5. We extend each 2D RoI at a frame into a 3D RoI by replicating it along the temporal axis. We compute RoI features by mRoIAlign [10] spatially, and global average pooling temporally.

However, I noticed that:
https://github.com/facebookresearch/SlowFast/blob/3a8b0bd1f5de90bc843b694eaa0cc7d845640f42/slowfast/models/head_helper.py#L73-L80

RoI Align is used before the temporal average pool and no extension for 2D RoI to 3D.