Many of you have probably seen the turtle from LabSix that gets mistaken for a rifle in Google's InceptionV3 image classifier. I read the paper and I understand how they apply EOT to 2d images and on the individual pixel values, but I am still unsure how they implement the EOT algorithm to the 3d model.
- Are they using EOT to perturb the individual coordinates in the 3d model's mesh? Or are they perturbing images of a turtle and then printing the turtle from the images?
- How do they check the InceptionV3 output iteratively without having to 3d print the object each time and check the probabilities given?
Any examples that someone can point to would also be very helpful.