Related to ImageNet-R dataset

[JAYATEJAK]

Hi @GengDavid ,

Currently, I am trying to replicate the results of the SLCA approach. I have a question regarding the ImageNet-R dataset and would appreciate your assistance.

In the paper, it is mentioned that the ImageNet-R dataset contains 200-class images, split into 24,000 and 6,000 images for training and testing, with a similar ratio for each class. However, the dataset appears to be in a single folder without a clear separation into training and validation sets.

I am kindly requesting that you provide the script or method used to split the ImageNet-R dataset into the designated training and validation sets as used in your experiments. Having access to this script or procedure would greatly assist me in replicating the experiments in your work accurately.

Thank you for your time and attention. I look forward to your response.

[GengDavid]

Hi, @JAYATEJAK ,

Thanks for your interest in our work! The ImageNet-R dataset for continual learning is proposed in DualPrompt, and you can refer to the official repo for details.

The dataset is split for training and testing with a ratio of 0.8. For a ready-to-use script in PyTorch for splitting the dataset, you can refer to this repo.