Open ml-illustrated opened 4 years ago
@ml-illustrated , Thank you for your nice input, I like your modification a lot!
Just wonder would it be possible to make it compatible with both torchvision 0.4.2 and previous torchvision which still uses train_data
stuff, e.g. torchvision 0.2.1?
Hi Yonglong,
Sure, I'll take a look and test the changes against at torchvision 0.2.1 and update the PR once my changes can support both.
Gerald
On Fri, Jan 3, 2020 at 3:42 PM Yonglong Tian notifications@github.com wrote:
@ml-illustrated https://github.com/ml-illustrated , Thank you for your nice input, I like your modification pretty much!
Just wonder would it be possible to make it compatible with both torchvision 0.4.2 and previous torchvision which still uses train_data stuff, e.g. torchvision 0.2.1?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/HobbitLong/RepDistiller/pull/14?email_source=notifications&email_token=AIOT564T7VKMHVXREOUARKLQ37EOTA5CNFSM4KCRDFQ2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEICKVFI#issuecomment-570731157, or unsubscribe https://github.com/notifications/unsubscribe-auth/AIOT566FRUDEUVATR64RPSTQ37EOTANCNFSM4KCRDFQQ .
Hello, fan of your great work. In trying to reproduce your CIFAR100 results via README example commands, they would end up in an error, e.g.,
A simple change is to add four properties to a new class
CIFAR100BackCompat
to minimize the code changes, as submitted in this pull request.For additional background, here's the Torchvision's MNIST source for reference: https://pytorch.org/docs/stable/_modules/torchvision/datasets/mnist.html#MNIST
relevant section:
Tested the changes with: