close
The Wayback Machine - https://web.archive.org/web/20220518190757/https://github.com/albumentations-team/albumentations/issues/477
Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add new transforms from https://arxiv.org/abs/1903.12261 #477

Open
3 of 7 tasks
ternaus opened this issue Nov 26, 2019 · 8 comments
Open
3 of 7 tasks

Add new transforms from https://arxiv.org/abs/1903.12261 #477

ternaus opened this issue Nov 26, 2019 · 8 comments
Labels
enhancement good first issue

Comments

@ternaus
Copy link
Collaborator

@ternaus ternaus commented Nov 26, 2019 •
edited

There is a set of Pixel Level transforms that is used in the work Benchmarking Neural Network Robustness to Common Corruptions and Perturbations

The authors also share the code => we can absorb some transforms that they have into the library.

https://github.com/hendrycks/robustness/blob/master/ImageNet-C/create_c/make_imagenet_c.py

  • ShotNoise
  • Defocus
  • GlassBlur
  • Zoom Blur
  • Pixelate
  • Frost
  • Spatter
@ternaus ternaus added enhancement good first issue labels Nov 26, 2019
@ternaus
Copy link
Collaborator Author

@ternaus ternaus commented Nov 27, 2019 •
edited

Pixelate in the paper == Dowscale with cv2.INTER_NEAREST in the library.

@ternaus
Copy link
Collaborator Author

@ternaus ternaus commented Dec 4, 2019

ShotNoise in paper == ISONoise in the Library

@akarsakov
Copy link
Contributor

@akarsakov akarsakov commented Mar 3, 2020

@ternaus I'd like to take up this enhancement.

@akarsakov akarsakov mentioned this issue Mar 6, 2020
Open
@akarsakov
Copy link
Contributor

@akarsakov akarsakov commented Mar 10, 2020

Hi @ternaus,

Frost transform requires precalculated images with ice crystals.
What's the best way of storing this images in albumentations code (all images require ~2MB)?

@Dipet
Copy link
Collaborator

@Dipet Dipet commented Mar 10, 2020

It seems that in this transformation we can use any image templates.. I think it's best to name this transformation as a TemplateTranform. And give the opportunity to set any image as a template.

@akarsakov
Copy link
Contributor

@akarsakov akarsakov commented Mar 12, 2020

That's great point!
Moreover looks like it would be better to allow setting list of desired images as template and pick the random one.

@akarsakov akarsakov mentioned this issue Mar 23, 2020
Merged
@akarsakov akarsakov mentioned this issue Mar 25, 2020
Open
@p4rallax
Copy link

@p4rallax p4rallax commented Sep 12, 2021

Has this been finished?

@akarsakov
Copy link
Contributor

@akarsakov akarsakov commented Sep 13, 2021

@p4rallax The code is finished (please see PRs mentioned in this issue) but it haven't merged to master yet.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
enhancement good first issue
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
4 participants
@ternaus @akarsakov @Dipet @p4rallax