Dali: Custom augmentation. How can I write a mean augmentation?

[bamfpga]

Hello,
I'd like to create a custom augmentation like this: input/mean(input)
I started with the mean computation. Is there a way I can achieve this by combining cupy.ElementWiseKernel and cupy.ReductionKernel. Is there a simpler way to achieve this or do I have to write some cuda kernels like this: https://github.com/NVIDIA/DALI/blob/master/dali/kernels/reduce/mean_gpu.cu? Do I stand a chance?
Thanks!

[JanuszL]

Hi,
This sounds like normalization. Please check this tutorial how to do that.

[bamfpga]

HI, I don't want to divide by standard deviation, but the mean. Can I compute the mean separately somehow?

[JanuszL]

Oh, I missed that you want to divide, not subtract. We are developing a mean reduce operator and then you can just divide the input by the result of the mean operator. We will get back to you when the relevant operator is ready.
@awolant - FYI

[JanuszL]

This PR should do what you are asking for. https://github.com/NVIDIA/DALI/pull/2395. Please check the nightly DALI build after it is merged.

[JanuszL]

@bamfpga - https://github.com/NVIDIA/DALI/pull/2395 just have been merged. Can you check the most recent nightly build that follows it to see if that works for you?

[bamfpga]

That's great! Thank you very much!

[JanuszL]

DALI 0.28 has been released, it should address this issue.

[bamfpga]

Great news. thank you!