search

RUB-SysSec / GANDCTAnalysis

Code for the ICML 2020 paper: Leveraging Frequency Analysis for Deep Fake Image Recognition.
MIT License
176 stars 24 forks source link

about data #29

Open Mrli6 opened 1 year ago

Mrli6 commented 1 year ago

when i run the code python run_classifier.py \submission_models\source_identification\celebA\cnn_pixel ~\datasets\celeba128 -s 1000 the model outputs the belowed results 45.10% of the images are from class 0 (451) 53.20% of the images are from class 1 (532) 0.40% of the images are from class 2 (4) 0.20% of the images are from class 3 (2) 1.10% of the images are from class 4 (11) what do the results mean?

Mrli6 commented 1 year ago

Another question, when i run the same code in the mode of run and debug, the results are so different.