Strong Baselines

<!-- [OTHERS] -->

We train Mask R-CNN with large-scale jitter and longer schedule as strong baselines. The modifications follow those in Detectron2.

Results and Models

Backbone	Style	Lr schd	Mem (GB)	Inf time (fps)	box AP	mask AP	Config	Download
R-50-FPN	pytorch	50e					config	model | log
R-50-FPN	pytorch	100e					config	model | log
R-50-FPN	caffe	100e			44.7	40.4	config	model | log
R-50-FPN	caffe	400e					config	model | log

Notice

When using large-scale jittering, there are sometimes empty proposals in the box and mask heads during training. This requires MMSyncBN that allows empty tensors. Therefore, please use mmcv-full>=1.3.14 to train models supported in this directory.