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

some questions about the data #15

Open luan-g opened 2 years ago

luan-g commented 2 years ago

Hi, I have some questions about the data in the image classification track.

  1. Do images in iid test and nuisance need to be cropped by bounding box？
  2. Can we use the context out of the bounding box when training and testing?
  3. And can we use the additional information in the 'csv' file like azimuth during training?
  4. About the test-time training, can we use the test data multiple times? Is there any constraint？
DTennant commented 2 years ago

Hi, Thanks for your interest for our challenge. Regarding your questions:

  1. We have prepared a data processing script and the processed dataset which will be released today or tomorrow, and yes, you need to crop the box from the image to do testing.
  2. We do not post any restrictions on how you use the data to train or test your model, the only restriction is that we penalize the final iid test performance, the details can be found on the website http://www.ood-cv.org/challenge.html
  3. Yes, and also azimuth is the target for 3D pose estimation.
  4. Yes, you can; the only constraint is the iid performance, as said earlier.