Skip to content

Potential Bug: inconsistent pixel value range for tiny-imagenet models #1

@njuaplusplus

Description

@njuaplusplus

The pixel range of ULP in tiny-imagenet/train_ULP.py is [0,255].

However, when training the clean/poisoned models, the pixel range of inputs is [0, 1] because of the transforms.ToTensor() in tiny-imagenet/train_clean_model.py

See transforms.ToTensor doc.

I wonder if the inconsistency is made purposely or is a bug. Because for CIFAR-10 experiments, the range is the same [0, 255].

By the way, I also wonder why the range [0, 255] was used instead of the usual [0, 1]. Are there any particular reasons?

Thanks.

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