Is it possible to make augmenter deterministic? #963
Unanswered
CesarHiersemann
asked this question in
Q&A
Replies: 1 comment
-
I was able to find a workaround for now (that works with the RandAugment layer atleast) by concating the two images in the color channel before running it through the layer:
The behavior I am ideally looking for (same as how it can be used in
|
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi KerasCV team,
Big fan of your work and this repo!
Quick question for you, is there a way to freeze/make_deterministic a (set of) augmenters so that the augmentation can be applied identically to two samples? Something similar to this imgaug functionality which is what I am using today, but I wish to make a full transition to tf.data preprocessing.
I am working with pairwise image inputs and wish to apply the same augmentation to each sample pair. I attempted to achieve it through equally seeding two instantiations of the same set of augmenters, but surprisingly the results are not consistent at all and definitely does not produce reproducible pairs of outputs. Even just seeding one augmenter and running over the same input does not give predictable results. Any insights into the seeding implementation here and if I am doing anything wrong?
Thanks for any input!
Beta Was this translation helpful? Give feedback.
All reactions