data structure

[ndavoudi]

Could you please share the data format. How the images and labels (with multiple landmarks) are saved in the directory.

[reza-asad]

Hi ndavoudi, could you share your email? I can send you the data.

[ndavoudi]

Hi Reza, thank you very much for your reply. This is my email address: neda.davoudi@tum.de I could manage to run your code on my data which is medical, I have a few questions:

• Regarding scaling the data: I understand normalizing the image by dividing to 255 but why do you multiply it by 2? And why do you divide landmark coordination by 255? Doesn't that make it inaccurate? especially if you have done resizing the image as well? Maybe having a look on your data could help.
• Your code was pretty intuitive and I could follow as my first experience with Torch and adapt it to my problem quite easy, do you by any chance have the script for 3D landmark detection as well?

Cheers Neda

[reza-asad]

Hi Neda,

I have sent you the data. Regarding your question on multiplying by 2: Usually you can get a better performance on your model if you normalize the images between -1 and 1. To do this I do 2 * x/255 - 1 which maps the pixels to be between -1 and 1. About your second question, unfortunately I have not done 3D landmark detection.

Cheers, Reza.