snu-mllab / efficient-dataset-condensation Goto Github PK

Hi, could you please tell how long does it take to condense ImageNet subset?

In your code, you just use it_test = [n_iter // 10, n_iter // 5, n_iter // 2, n_iter] to evaluate the test accuracy of condensed data. However, when I rerun your code, I didn't get ideal result that described in your paper? So will you use more frequent evaluation in your training. And will you calculate this evaluation time in your table?

Thank you very much for your sincere answer.

Hello, Thanks for your solid work.
When I want to reproduce your code in imagenet100 ipc=20, I use the command of python condense_mp.py --reproduce -d imagenet --nclass 100 --pt_from 5 -f 1 --ipc 10 --nclass_sub 20 --phase 5, I found that there are some error in running it.
So how many GPUs did you use in your experiment, and how long will it take?

Hello, first of all, congratulations on the work.

I'm trying to reproduce the results of speech signals and have some questions.

1)It seems that only spectrograms were used as the audio representation, am I correct? Was there any testing done with mel spectrograms?
2)Once you have obtained the condensed spectrograms, how do you return to the time domain? Is it necessary to perform the decoding?How does it work?

Thanks in advance

Hello, I want to reproduce your result on Imagenet10 and imagenet100. However there is a question for that there are so many kind of Imagenet10 and imagenet100. So can you please give us a official link to get the Imagenet10 and imagenet100, which is the same as yours.
Thank you for your sincere help.