"one training dataset"

[Balázs Knakker]

Hi!

In the description, you mention the following:

"It uses a unique data augmentation strategy combines visual simulation and negative output saturation to achieve accurate and efficient segmentation with just one training dataset. "

What exactly do you mean by "just one training dataset"?

Thanks!

Balázs

[Fang-Cheng Yeh]

For rhesus, I used only one T1W image from ONPRC18 publicly available
data from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7833476/
That is all it took to train the network.
Frank

> --
> You received this message because you are subscribed to the Google Groups "UNet Studio" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to unet-studio...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/unet-studio/bdb7de5e-9d2e-4830-98fd-2e1478650dd7n%40googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[jongsung park]

This is very interesting! I would love to hear how the augmentations were done to have this result. Is there a paper in review or being written?

2023년 4월 24일 월요일 오후 12시 48분 39초 UTC-4에 Fang-Cheng Yeh님이 작성:

[Fang-Cheng Yeh]

Not yet, there is still room for improvement and would need more user
testing to see the issues.

> To view this discussion on the web visit https://groups.google.com/d/msgid/unet-studio/6050573c-d56e-4cd3-afdc-7eca0b657ed2n%40googlegroups.com.