Skip to content

Just get a low accuracy of 75.8 with resnet50 on ImageNet #1153

@mountain111

Description

@mountain111

I train resnet50 on ImageNet with GPUs=8, batchsize=256, learning-rate=0.1, epochs=90, and momentum=0.90.
The attained top1 accuracy is 75.80, lower than the reported 76.15. The gap is not marginal on the large-scale ImageNet.
Why does the difference exist?

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