search

mdbloice / Augmentor

Image augmentation library in Python for machine learning.
https://augmentor.readthedocs.io/en/stable
MIT License
5.07k stars 866 forks source link

Possible Improvment #103

Open tchaton opened 6 years ago

tchaton commented 6 years ago

Good morning,

I have been playing a lot with Augmentor, and I have several suggestions to really improve it. 1 . Add possible npy loading (I have almost finished implementing it) 2 . Rely more on https://github.com/aleju/imgaug for imgAug. (We don t have to re-implement everything) 3 . Add support for random variable into the pipeline, each element could be sample and improve randomness from the data augmentation.

Best Regards, Thomas Chaton.

jsherman256 commented 6 years ago

I would love it if I could pass a DataFrame to the Pipeline constructor. My training set is 28x28 pixel images in a DataFrame (not actual image files). It would be very convenient to just pass that DataFrame directly to the constructor. Is that what you are implementing as part 1? If so, I'd be very interested in using it :)

tchaton commented 6 years ago

Using the data frame pipeline U can. U just need to have file path and label .

Le dimanche 13 mai 2018, jsherman256 notifications@github.com a écrit :

I would love it if I could pass a DataFrame to the Pipeline constructor. My training set is 28x28 pixel images in a DataFrame (not actual image files). It would be very convenient to just pass that DataFrame directly to the constructor. Is that what you are implementing as part 1? If so, I'd be very interested in using it :)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/mdbloice/Augmentor/issues/103#issuecomment-388609188, or mute the thread https://github.com/notifications/unsubscribe-auth/AMRCHdifUYaRN3dpmliV_0Phh5cjf0VWks5tx-nrgaJpZM4TMG2V .