search

mdbloice / Augmentor

Image augmentation library in Python for machine learning.
https://augmentor.readthedocs.io/en/stable
MIT License
5.06k stars 865 forks source link

Add support for semantic segmentation problems #51

Open kmader opened 6 years ago

kmader commented 6 years ago

It would be useful if the pipelines could be applied in parallel to both the image and label dataset for segmentation problems. It is possible to sync two pipelines and have them operate independently on the input images and output labels, but it would be useful to support both

Evizero commented 6 years ago

Hi! This feature is definitely on our radar, but currently not implemented natively.

However it seems there is a (somewhat errorprone) workaround available. see #39

yanmu2017 commented 6 years ago

I tried #39 , but I have problems to obtain label images interpolated with 'NEAREST' interpolation. It seems I can choose parameter 'resample_filter' only for the method 'resize', but not others like 'rotate' and 'zoom'.

It would be very helpful if we can set a 'resample_filter' flag for a pipeline (Augmentor.Pipeline) that is valid for all its augmentation operations.

mdbloice commented 6 years ago

Hi @yanmu2017 that's a good idea, it would make sense for all operations to just use one resample filter for the entire pipeline. I will do that in the next version.

tmquan commented 6 years ago

May I ask whether the 'NEAREST' interpolation has supported yet? Is it fine to pass the interpolation strategy as an argument to the pipeline?