Is Mask-RCNN a fully convolutional architecture?

[justlike-prog]

Is Mask-RCNN a fully convolutional architecture (from what I see yes, because there are no dense layers but just wanted to make sure)? Thus can I feed differently sized images for inference without resizing? Also while training could I feed the original sized images without resizing?

[ppwwyyxx]

Thus can I feed differently sized images for inference without resizing

you can.

Also while training could I feed the original sized images without resizing?

you can, though this is not common