CUDA 11.2 is quite old now, is there any intention of supporting newer versions of CUDA? The main constraining factor is `maskrcnn_benchmark` which has been deprecated and replaced by `detectron2`.