Skip to content

复现结果相差非常大 #166

@xlyyydGH

Description

@xlyyydGH

要测试 Celeb-DF-v2跨数据集的效果时,验证集应该设置成一还是二呢?如果是一auc一直很低怎么办?
train_dataset: [FaceForensics++]
test_dataset: [FaceShifter,Celeb-DF-v2]

train_dataset: [FaceForensics++]
test_dataset: [FaceForensics++,FaceShifter,Celeb-DF-v2]
2.我用您提供的预训练权重进行测试,只得到了以下的结果,跟论文里相差非常大,我自己训练的话在Faceshifter只能达到50+的auc

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions