GeorgeCazenavette / mtt-distillation

Official code for our CVPR '22 paper "Dataset Distillation by Matching Training Trajectories"
https://georgecazenavette.github.io/mtt-distillation/
Other
401 stars 53 forks source link

CIFAR 10 ipc 10 Hyperparameters #31

Open WeizhiGao opened 1 year ago

WeizhiGao commented 1 year ago

Hi George,

Thanks for your amazing work!

I am testing your codes in the setting of CIFAR10 ipc 10. I adopt the hyperparameters list in the readme page. However, gradient explosion ocurrs on parameter 'syn_lr'. By changing the learning rate (pxiel) to 100 or 1000, I can get a stable result. I wonder if there is a typo in the table.

Your anwer will be really helpful!

Much Thanks.

Byunggwan-Son commented 1 year ago

I am having the same problem.

GeorgeCazenavette commented 1 year ago

Hi all, apologies for the late reply.

Yes, this is definitely a typo in the table, sorry about that.

I can try to dig up the old experiment logs and confirm the correct value we used if you're curious.

Thats2Great commented 1 month ago

I have the same problem