The Dude
The Dude
@fsx950223 do u sample objects from same batch or other batches?
I have implemented this copy and paste in my local repo will file a pr or may be @fsx950223 can file it too :)
Yes will submit soon :)
You sure can treat bbox as objects and copy them around but it can add background noise from the bbox
@fsx950223 @mingxingtan Copy and Past augmentation itself is supervised learning but the author combined this technique with self training method like psuedo labels hence made the entire trainging semi-supervised. but...
@fsx950223 copy and paste function itself is easy like Def copy(image1 objects, image2objects): . . Challenge is to sample these masks/objects in our input pipeline implementation If we could sample...
Well then i will refer your codes above and will file pr, but you have to help to install the code in our input pipeline.
Hi, The idea efficientNetV2 is centered around is progressive learning and also ( architectural changes), hence by just changing backbone won't be enough, progressive image learning should be added in...
> Yeah, you should be easily plugin efficientnetv2 to efficientdet by changing this line: https://github.com/google/automl/blob/d07468a28df7100251332865332f2412eafe6f31/efficientdet/keras/efficientdet_keras.py#L805 > > > > Hopefully I will get TPU resources soon, and retrain those detection...
> Oh, no progressive learning, I meant just the backbone. > > > > Never tried progressive learning for detection, but in general, the anchor boxes might be the issue....