Closed solucky-95 closed 3 years ago
Hello,
Thank you for your time. You have provided two pre-trained models for cifar100 and imagenet-subset respectively. The cifar100 one can perfectly reproduce your results, but the results from the imagenet-subset one are quite low (for task 0, it only gets 30%-40%). Are you using the first 100 classes of the original imagenet or applying random seed 1993 and then selecting the first 100 classes? I have tried both but the results are not good. From my understanding, the starting point (task 0) should produce the same results using your pre-trained model.
Best.
Hi, would you please share your results on CIFAR? Since there is no accurate value in the original paper, is would be really nice to have your results.
Hi @solucky-95 ! Did you find the reason for the poor performance on Mini-Imagenet dataset? If yes it would be great if you can share what was the problem and how did you solve it.
@yulu0724 May you help and provide some detailed instructions on how to reproduce the results for Mini-Imagenet dataset?
Hello,
Thank you for your time. You have provided two pre-trained models for cifar100 and imagenet-subset respectively. The cifar100 one can perfectly reproduce your results, but the results from the imagenet-subset one are quite low (for task 0, it only gets 30%-40%). Are you using the first 100 classes of the original imagenet or applying random seed 1993 and then selecting the first 100 classes? I have tried both but the results are not good. From my understanding, the starting point (task 0) should produce the same results using your pre-trained model.
Best.