Prediction script for 3D volumes

[kiranvaidhya]

Hi,

Could you provide the prediction scripts you use for iterating and predicting through 3D volumes? I'm getting prediction artifacts between sub-volumes because of padding present inside the network. Would be great if you can provide solutions for it.

[faustomilletari]

Hello, the code includes all the necessary to do prediction of a single volume, without using overlapping patches (all is segmented at once). people have reported problems with this which looked like padding artifacts. in reality these were not padding artifacts but were issues related to the fact that when you select the resolution of your input volumes in the main file you have also to select the final size of the input volumes which is independent from the resolution. sometimes this results in cropping parts of your volume and therefore cropping anatomies in half. the final segmentation will not exhibit this cropping as a new volume of the same size of the input volume will be given back to you as a result. if you want to have very high resolution you can do an overlapping patch approach. the code for this is a bit tricky to write and is not provided by me.

…

On May 22, 2017, at 12:08 PM, Kiran Vaidhya ***@***.***> wrote: Hi, Could you provide the prediction scripts you use for iterating and predicting through 3D volumes? I'm getting prediction artifacts between sub-volumes because of padding present inside the network. Would be great if you can provide solutions for it. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#39>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMtsvvgJdTQNj_J0WZ097efwabZERu_cks5r8bLqgaJpZM4NikXG>.

[abhijithchunduru]

Hi,
I am also facing the same problem. I did not understand your explanation. Can you please elaborate ?

[faustomilletari]

There was somewhere another issue by somebody having the same problem. The thing is if you start from a volume that is 1000x1000 at 1 mm per pixel, and you use default settings, the system will crop out a 128x128x64 crop and elaborate that part only! When you get your results back you get a 1000x1000 where you will see a sharp cut in the segmentation due to the fact that only a 128x128x64 portion was seen. Fausto Milletarì Sent from my iPhone

…

On 22. May 2017, at 12:40, Abhijith Chunduru ***@***.***> wrote: Hi, I am also facing the same problem. I did not understand your explanation. Can you please elaborate ? — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or mute the thread.