Skip to content

Pruning Problem of MobileNetV2 for CIFAR100 dataset #11

@tahsirmunna

Description

@tahsirmunna

Hello @RudyChin ,

Thank you for your brilliant work.

I am facing some problems when I used the CIFAR100 dataset. For CIFAR10 it works well. I got the same result that you showed in the paper for CIFAR10.

But for CIFAR100 dataset, When I select 0.9 pruning away for 60 epochs even 1000 epochs, it just prunes till 80% flops where targeted flops should be 26.58M for 90% prunes. I tried in a multiple-way to change the hyper-parameters but it does not work.

Would you please, give me some suggestions that how I can solve this problem?

Here are my used hyper-parameters for your reference,

Namespace(datapath='./data', dataset='torchvision.datasets.CIFAR100', epoch=1000, name='prune_90', model='./mbnetv2c100-best.pth', batch_size=128, lr=0.01, lbda=3e-09, prune_away=0.9, constraint='flops', large_input=False, no_grow=False, pruner='FilterPrunerMBNetV2')

Thank you and eagerly waiting for your valuable reply.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions