search

AdelehBitarafan / Vol2Flow

8 stars 0 forks source link

Data Preprocessing & Train time #1

Open RachaellNihalaani opened 8 months ago

RachaellNihalaani commented 8 months ago

Hello,

I am trying to use your code, the same datasets mentioned in the paper. From your code (train.py), I understand that it is expecting .npy files in a dictionary format? Could you please elaborate on what exactly you do to convert datasets from NifTi or other formats to .npy?

AdelehBitarafan commented 8 months ago

Hello, I have used pydicom library for that. You can read dicom files using dcmread function.

RachaellNihalaani commented 8 months ago

Thank you. I had to modify the Data Loader a bit, but managed to make it work.

Could you tell me how long did it take you to train your models? It is taking me 18+ hours for a single epoch and I am wondering if that is expected or if I am missing out on something?

AdelehBitarafan commented 8 months ago

For me, each epoch takes about 3 to 4 hours if the input dimension be 128*128*400. However, you can decrease it by changing the input dimension. For example, if the input be 128*128*64, the model can be trained faster.

Let me know if you have another question.