NaN or Inf in lung_nodule_ct_detection during inference

[izvarinanat]

Hello!
I try to run lung nodule detection model inference with my own data.
I prepared my dicoms in accordance to tutorials/detection/luna16_prepare_images_dicom.py and run monai.bundle inference.
.pt model was taken from https://developer.download.nvidia.com/assets/Clara/monai/tutorials/model_zoo/model_lung_nodule_ct_detection.pt
There weren't any changes in inference.json file except paths to dataset directory, annotations and .pt model.
During inference I catch this error:

cls_logits is NaN or Inf.
box_regression is NaN or Inf.
Concatenated result is NaN or Inf.

Debugging shows that all model outputs (classification and box_regression) are nan.
Could you help me to fix this problem.