search

ZFTurbo / Keras-RetinaNet-for-Open-Images-Challenge-2018

Code for 15th place in Kaggle Google AI Open Images - Object Detection Track
MIT License
268 stars 76 forks source link

How can I add an color augmentation into the transform.py? #12

Open ZenFSheng opened 5 years ago

ZenFSheng commented 5 years ago

Hi~ I decide to use your model as my pretrained model，but I‘ve found that there's no color augmentation in the generator，and it seems that the keras-retinanet use a modified transform generator， can you help me with that？Thank you very much~

ZFTurbo commented 5 years ago

If you use generator from this repository, it already has some color augmentations on this line:

https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/retinanet_training_level_1/open_images_generator.py#L347

ZenFSheng commented 5 years ago

If you use generator from this repository, it already has some color augmentations on this line:

https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/retinanet_training_level_1/open_images_generator.py#L347

Thanks a lot~，I also notice that there are some morphological augmentation in the generator，is it work well for your training ？

ZFTurbo commented 5 years ago

Yes. It's the exact augmentations I used previous year. It works fine. Morphological augmentation was used from original RetinaNet repository. This year I rewrote them to use albumentation library.

ZenFSheng commented 5 years ago

Yes. It's the exact augmentations I used previous year. It works fine. Morphological augmentation was used from original RetinaNet repository. This year I rewrote them to use albumentation library.

Thank you！That’s really helpful to me~