Size mismatch when running FasterRCNN in parallel

Detection models do not seem to support nn.DataParallel, so you would need to use nn.DistributedDataParallel instead (which we also recommend to use anyway).
Have a look at this post for more information.

CC @Anjum_Sayed