MONAI nodule detection model training and handling zero nodule datasets

[DSRajesh]

Hello
Thank you for the MONAI team. I have been training a nodule detection model using

Among my datasets I have some which have 0 nodules in them. And i think i get a error log as shown below for such cases

I have applied the correction mentioned there

i.e
boxes = torch.reshape ( boxes, [ 0, 4 ] ) as follows:

but i still get the same error log as shown above. May i know the reason for this and the solution for the same please. Also I have used datasets that have 0 nodules and a nnDetection (https://github.com/MIC-DKFZ/nnDetection) model trains well under such situations.

Rajesh D S

[KumoLiu]

Hi @DSRajesh, could you please refer whether the discussion here helps you? Thanks.

[DSRajesh]

Hello @KumoLiu
Thank you for the response. The discussion you pointed out me to in your response discusses the same solution that i had already tried as follows:

and had still resulted in error.

Also the discussion (Project-MONAI/tutorials#1197) was with reference to MONAI tutorials for "object detection" (https://github.com/Project-MONAI/tutorials/tree/main/detection). I am using the MONAI zoo model for nodule detection i.e

.

Do I need to add some lines in the config/train.json file to incorporate the corrections mentioned in Project-MONAI/tutorials#1197 ?

Thank you
Rajesh D S