Closed zhaoedf closed 1 year ago
Hi,
As mentioned in https://github.com/google-research/l2p/issues/27, I map class labels to 0-199 and then split in the natural order, i.e., 0-19, 20-39, etc.
A better way is to try different splits (by playing with the seeds) and report the average performance.
Best, Zifeng
i was wondering if you could provide the seed and the corresponding class orders used in your imagenet-r experiment?