search

eccv22-ood-workshop / ROBIN-dataset

ECCV 2022 Workshop: A Benchmark for Robustness to Individual Nuisances in Real-World Out-of-Distribution Shifts
http://www.ood-cv.org/
46 stars 6 forks source link

Clarifications on Image Classification Track #28

Closed PRE6ENT closed 2 years ago

PRE6ENT commented 2 years ago

Hi,

Thank you for hosting such a workshop on OOD.

I have some general questions regarding the image classification track.

1) I thought the model for training / testing was restricted to ResNet50 only, but I can't find this restriction now. Did the rules change?

2) For ensemble training, are we allowed to use models other than ResNet50?

3) For the description at the end of Phase 1, how are we supposed to upload the description file?

Thank you in advance.

DTennant commented 2 years ago

Hi:

Regarding your questions:

  1. We do not have the restriction of ResNet50 only, our restrictions are, that the model can only be trained on ImageNet1k and/or our provided training set, and also the IID test performance should not exceed a predefined threshold.
  2. Yes, you can use any model for ensembling.
  3. We will send out emails to all teams on CodaLab at the end of phase 1 and ask for the description file.

CJaehyun @.***> 于2022年9月16日周五 00:44写道：

Hi,

Thank you for hosting such a workshop on OOD.

I have some general questions regarding the image classification track.

1.

I thought the model for training / testing was restricted to ResNet50 only, but I can't find this restriction now. Did the rules change? 2.

For ensemble training, are we allowed to use models other than ResNet50? 3.

For the description at the end of Phase 1, how are we supposed to upload the description file?

Thank you in advance.

— Reply to this email directly, view it on GitHub https://github.com/eccv22-ood-workshop/ROBIN-dataset/issues/28, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADY4MVOLP4IIBMUPPAFSLS3V6NG55ANCNFSM6AAAAAAQNSLZN4 . You are receiving this because you are subscribed to this thread.Message ID: @.***>