Why evaluates on cpu

[9p15p]

mmengine/mmengine/evaluator/metric.py

Line 132 in eb4fa73

results = _to_cpu(results)

it runs slowly on cpu. If I want to use GPU to speed up the evaluation. What should I do?

[YiyaoYang1]

Sorry, these calculations cannot be accelerated with GPU. In this stage, the sample category labels and classification predictions are extracted from the results to calculate the classification accuracy (acc). This calculation can only be performed on CPU.

[9p15p]

Thanks for your reply.

I am building my project(not classification) based on mmengine
I think it's a good idea to give users an option to complete time-consuming measurements on GPU.