Skip to content

Reproducing table 1 results #2

Description

@kh12043

Hello, and thank you for releasing the code for your impressive work.

When I ran your code, I obtained results that differ somewhat from those reported in the paper. Specifically, when reusing a single model as in Table 1, I obtained the following results:
(I used the virtual environment provided in your repository.)

cifar100 & country211 & clevr-d & dtd & dmlab & flowers102 & mnist & oxfordpet & pcam & fer2013 & food101 & gtsrb & resisc45 & rendered-sst2 & stanfordcars & stl10 & ucf101 & Avg.

0.7883 & 0.2522 & 0.1997 & 0.6447 & 0.1514 & 0.8059 & 0.8450 & 0.9381 & 0.5002 & 0.2357 & 0.9253 & 0.4380 & 0.6167 & 0.5200 & 0.9120 & 0.9945 & 0.7772 & 0.6205

It seems there are several potential sources of randomness, such as during sampling dataset construction and caption generation. Could you let me know which random seed you used in your experiments?

Additionally, I would appreciate any guidance that could help with reproducing the results reported in the paper.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions