Skip to content

How to generate FP16 CLIP and other metric results in text-to-image experiments? #21

@xululi0223

Description

@xululi0223

Thank you for the recently released evaluation code. When evaluating on the COCO dataset, it seems that the CLIP scores and other metrics are obtained by comparing images generated from the quantized model with those from the full-precision model. In that case, how can we compute the CLIP scores of the original FP16 model as reported in the paper? Thank you!

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