Could you explain the difference between the binary search in the initialize method [1] and the binary search right after it [2]? It seems to me that the second one is unnecessary, since the image already has been projected to the boundary.
[1] Binary search in the initialize method
|
# Binary search to minimize l2 distance to original image. |
|
low = 0.0 |
|
high = 1.0 |
|
while high - low > 0.001: |
|
mid = (high + low) / 2.0 |
|
blended = (1 - mid) * sample + mid * random_noise |
|
success = decision_function(model, blended[None], params) |
|
if success: |
|
high = mid |
|
else: |
|
low = mid |
|
|
|
initialization = (1 - high) * sample + high * random_noise |
[2] Binary search right after initialize
|
# Project the initialization to the boundary. |
|
perturbed, dist_post_update = binary_search_batch(sample, |
|
np.expand_dims(perturbed, 0), |
|
model, |
|
params) |
|
dist = compute_distance(perturbed, sample, constraint) |