search

Jingkang50 / OpenOOD

Benchmarking Generalized Out-of-Distribution Detection
MIT License
831 stars 106 forks source link

Missing file #241

Open gugu777 opened 3 months ago

gugu777 commented 3 months ago

Hi！ I couldn't find the file in the download link/ data/benchmark_imglist/cifar10/val_cinic10.txt， If possible, could you send me a copy

zjysteven commented 3 months ago

The v1 benchmark_imglist files can be found here, however I didn't see a val_cinic10.txt. Even if it were presented in v1 it definitely is removed in v1.5.

Jingkang50 commented 3 months ago

Lemme try to find it.

Jingkang50 commented 3 months ago

I thought we were following the data here https://paperswithcode.com/dataset/cinic-10 to get the list. So if I could not make it to find the txt, please simply use the original data split

gugu777 commented 3 months ago

The v1 benchmark_imglist files can be found here, however I didn't see a val_cinic10.txt. Even if it were presented in v1 it definitely is removed in v1.5. However, the dataset partition file appears in the script cifar10_test_ood_mem.sh. Additionally, by default, I am unable to reproduce the experimental results in the paper. Could you please send me the correct parameters? Thank you very much.

zjysteven commented 3 months ago

Which exact script are you referring to? Is it cifar10_test_ood_sem.sh?

Also, by "paper" do you mean the v1 or v1.5 paper? A few datasets and methods presented in v1 are not considered in v1.5 (since v1.5 focuses more on the large-scale settings).

gugu777 commented 3 months ago

Which exact script are you referring to? Is it cifar10_test_ood_sem.sh?

Also, by "paper" do you mean the v1 or v1.5 paper? A few datasets and methods presented in v1 are not considered in v1.5 (since v1.5 focuses more on the large-scale settings).

I am referring to the paper "Full Spectrum Out of Distribution Detection".

zjysteven commented 3 months ago

Then Jingkang might have better idea on that.

Jingkang50 commented 3 months ago

The parameter used for MNIST should be -1, 0, 1. The parameter for CIFAR should be smaller, like -0.01, 0, 1

gugu777 commented 1 month ago

Hi！
I would like to ask about the setting of CSID, for example, for ID cifar10, its CSID is written as cifar10-c and imagenet-10 in the paper, but in the configuration file in the code, only cifar10-c is written. Do I need to manually add another one or just keep cifar10-c.