Closed manthansinghshekhawat closed 1 year ago
This is a good question. But unfortunately, I haven't tried that before. Our codebase is built on top of Detectron2 and thus I'd recommend searching the usage of Detectron2. One quick solution is simply removing all .cuda()
in the data loading and model forward, which port tensor onto GPU.
I am running extract_region_feature.py and it's showing torch is not compatible with cuda I don't have cuda enabled gpu can I run this code on cpu only