Interpreting Top5 Accuracy

MadryLab / robustness

A library for experimenting with, training and evaluating neural networks, with a focus on adversarial robustness.

MIT License

903 stars 181 forks source link

Interpreting Top5 Accuracy #110

Open alexriedel1 opened 2 years ago

alexriedel1 commented 2 years ago

Hey, I'm training a robst model. During training I get a AdvPrec5 of 1.369. How can this value be intrepreted? Doesn't the Top5 precision have to be in the range [0,1] ?

Train Epoch:0 | Loss 6.8375 | AdvPrec1 0.346 | AdvPrec5 1.369 | Reg term: 0.0 ||: 97%|█████████▋| 1936/2002 [1:14:17<02:31, 2.30s/it]