-
Notifications
You must be signed in to change notification settings - Fork 18.7k
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
data augmentation tool/feature #701
Comments
I have the same problem. I don't want to re-create a dataset with ~20 million images every time I change my augmentation strategy, so I implemented on the fly augmentation. I implemented an augmentation step which randomly perturbs images by modifying data_layer.cpp. Every time datum.ParseFromString appears in this file, I add my own function Perturb(&datum), which modifies the datum by distorting the image and resizing it (I removed the requirement that all images in the leveldb have the same size because I don't want to distort downsampled images). I think support for on the fly processing from the leveldb is sufficiently important to merit inclusion in caffe. Does anyone have any thoughts/suggestions on a clean way to implement this? |
I think that augmentation can share many things with "synthetic" dataset generation |
See also http://hal.inria.fr/hal-00979464/ |
Sander Dieleman implemented realtime augmentation as described in "My solution for the Galaxy Zoo challenge". |
Another interesting experiment on augmentation: http://infoscience.epfl.ch/record/199706 |
try multi-scale?see this paper:http://arxiv.org/abs/1406.2984 |
@shelhamer, four existing implementations haven't been taken care of for a long time: #1386, #1865, #1900, #2252. |
Please ask usage questions on the caffe-users list, and document related tools and techniques on the wiki. Thanks! |
@senecaur Are you interested to contribute something back with a PR from your fork? |
I was wondering what tools or features are available in caffe for typical data augmentation steps for image datasets?
The text was updated successfully, but these errors were encountered: