search

softsys4ai / athena

Athena: A Framework for Defending Machine Learning Systems Against Adversarial Attacks
https://softsys4ai.github.io/athena/
MIT License
42 stars 9 forks source link

poisson-noise transformation model crashes when predicting the AE, jsma (theta-10, gamma-30) #18

Closed oceank closed 4 years ago

oceank commented 5 years ago

[Error Message]

Traceback (most recent call last): File "detection_as_defense.py", line 67, in transformationList) File "/home/kevinsjh_gmail_com/adversarial_transformers/util.py", line 1087, in predictionForTest tranSamples = transform_images(curSamples, transformType) File "/home/kevinsjh_gmail_com/adversarial_transformers/transformation.py", line 748, in transform_images elif (transformation_type in TRANSFORMATION.NOISES): File "/home/kevinsjh_gmail_com/adversarial_transformers/transformation.py", line 671, in add_noise img_noised = skimage.util.random_noise(img, mode=noise_mode) File "/home/kevinsjh_gmail_com/anaconda3/lib/python3.7/site-packages/skimage/util/noise.py", line 155, in random_noise out = np.random.poisson(image * vals) / float(vals) File "mtrand.pyx", line 4005, in mtrand.RandomState.poisson ValueError: lam value too large.

MENG2010 commented 5 years ago

Not producible. Steps:

  1. load adversarial examples (test_AE-mnist-cnn-clean-jsma_theta10_gamma30-noise_poisson.npy).
  2. apply noise_poisson on the adversarial examples. noise_poisson was applied correctly.

Maybe these adversarial examples have been changed before you applied the transformation.

MENG2010 commented 4 years ago

fixed by clipping generated images into proper range.