search

rabi3elbeji / SteganalysisCNN

An Improved CNN Steganalysis Architecture Based on “Catalyst Kernels” and Transfer Learning
MIT License
25 stars 12 forks source link

some issue about result #4

Open iMhds opened 4 years ago

iMhds commented 4 years ago

Hi I trained 3400 images ( 1700 cover,1700 stego ) but I got same loss value for wow 0.3 and wow 1.0 ( LOSS = 0.6125 ) In predict phase I always get AUC = 0.5 There must be something wrong :( PS:

  1. Since ImageDataGenerator doesn't support pgm format ( dataset boss_base ) I changed pgm to png. does it make sense ?
  2. I use google colab Thank you
cuiqi369 commented 2 years ago

你现在解决了吗，不需要更改图像格式