Aakash Kumar Nain
Aakash Kumar Nain
@koshian2 excellent explanation. I have one question though. If I am correct, the `high` and `low` parts are chosen from the channel dimension with `1-alpha` and `alpha` respectively. If I...
@koshian2 Thanks for the explanation. That is much clear now. Also, can I use your implementation as a reference implementation for my work?
@WindQAQ Yes, that needs to be changed to `AUTO` and we need to make a few other changes as well. But I won't be able to fix it before next...
Same thought. It would be much better if this is implemented as a layer.
@TomAugspurger @mrocklin Once we have our `stacked` delayed dask array, can't we just generate batches of data from it on the fly? Something like this ```python data = [dd.array.from_delayed(x, shape=(224,224,...
Thanks @mrocklin I would elaborate a bit on that. Say I have `50,000` images on my disk. I cannot load all the data in memory once. In normal case, we...
Cool. Thanks a lot for your time. Yeah, I am aware of those issues, and that is why I just want to use `dask` for batch generation and no delayed...
@skeller88 I didn't try it. I was trying to benchmark it with tf.dataset. But this certainly looks good to this point.
Hey @colah . Thank you for the information. I will try it out and will let you know if it works the way as expected