buzhiqimeiliuqiangdong
commented
1 year ago Dear Authors,
I have a question regarding the results in Table 4 in your paper. T2T gives 99.85% AUROC on CIFAR-ID-50 with TIN. However, this setting does not have a hold-out test set for OOD classes because you use the test set of TIN for training. So on which test set do you compute the result 99.85%?
Using your code, I can only reproduce this number if evaluating the model on the training set (CIFAR-ID-50 unlabeled set + TIN). So my understanding is this number is training AUROC and should not be compared with other numbers.
兄你有这篇论文里测试AUROC的代码吗，作者的代码里好像没给
Dear Authors,
I have a question regarding the results in Table 4 in your paper. T2T gives 99.85% AUROC on CIFAR-ID-50 with TIN. However, this setting does not have a hold-out test set for OOD classes because you use the test set of TIN for training. So on which test set do you compute the result 99.85%?
Using your code, I can only reproduce this number if evaluating the model on the training set (CIFAR-ID-50 unlabeled set + TIN). So my understanding is this number is training AUROC and should not be compared with other numbers.