search

megvii-research / CADDM

Official implementation of ID-unaware Deepfake Detection Model
Apache License 2.0
158 stars 21 forks source link

bias in bbox #2

Open xarryon opened 1 year ago

xarryon commented 1 year ago

Thanks for this new deepfake detection work! I tested this project and find the bboxes may bring biases during generation. Does this phenomenon may harm the performance? image

Nku-cs-dsc commented 1 year ago

It may increase the bounding box regression loss during the training phase. In my experiments, such high regression loss will limit the softmax loss convergence. So I think It may harm the model performance, please check your code and fix this bug~

xarryon commented 1 year ago

Actually I just use the code from CADDM without much changes, have you ever encounter this phenomenon?