Transfer-Learning-Library
Transfer-Learning-Library copied to clipboard
Why no torchmetrics for computing test time accuracy?
I noticed that there is a deviation from standard practice, during test time, the accuracy metric is just using 'top1' accuracy and it is not getting averaged across classes!
In short, two questions:
-
Why only top1 during test time? (and then, what does "origin" means, does that also mean Top1 test accu.)
-
Why not use the metrics from pytorch?
Hey any updates on this? @phamour @microhhh @longmingsheng. @thucbx99 (you answered some of the questions, pls can you answer if you know)
Also, I wanted to ask is there a possibility in near future that ViT backbones are added?