search

jgkwak95 / AU-GAN

Official Tensorflow implementation for "Adverse Weather Image Translation with Asymmetric and Uncertainty aware GAN", BMVC2021
67 stars 9 forks source link

Saved image is NAN #15

Open H-deep opened 2 months ago

H-deep commented 2 months ago

Thank you for the provided code and the pre-trained weights. I ran your test code for bdd100k dataset. However, I got the output saved images as black images. I check the output of the model: fake_img, = self.sess.run([out_var], feed_dict={in_var: sample_image}) in AUGAN.py script and the result is all NAN list: fake_img, = self.sess.run([out_var], feed_dict={in_var: sample_image})

The sample image is as follows: sample_image: [[[[-0.85882354 -0.81960785 -0.7882353 ] [-0.85882354 -0.81960785 -0.7882353 ] [-0.85882354 -0.81960785 -0.7882353 ] ...

But fake_img is as follows: fake_img: [[[[nan nan nan] [nan nan nan] [nan nan nan] ...

So the output of the concatenation is a nan also.

What do you think is the source of the problem?

Thank you

jgkwak95 commented 1 month ago

Hi. Please check the resolution or scale of your input images.