Inference with another database about alveolar_canal HOT 4 CLOSED

Hi @HamidFsian, sorry for being late in responding but I missed the notification. Could you please provide a better description of "it fails each time"? Does the code crash with some specific error or did you get really low metrics?

from alveolar_canal.

Hi @LucaLumetti ,
Thank's for your response.
In fact, I wanted to test your work on another database which contains a CBCT volume of the whole skull. and each time I pass this whole volume to the network, it give me low metrics.

Do you think that a pre-processing is required, where I crop the volume and take only the lower jaw for the test ?

Please not that I changed the batch_size from 6 to 1, because I have not enough computation capabilities for it.

Thank you,
Hamid

from alveolar_canal.

@HamidFsian as you are using the whole skull, it might be that the network misclassifies the voxels of the skull that has never been seen during training. I would have to check the segmentation map that it outputs to understand it better.
To solve this issue, you could probably do a little fine-tuning using your dataset

from alveolar_canal.

I agree with @LucaLumetti, fine-tuning is the best option. If you have no annotations on your volumes you can use our annotation tool. If annotating new CBCT is not an option you can still crop your data before feeding the network.

from alveolar_canal.