The preprocessing transforms in train.json and inference.json do not match for whole body ct segmentation bundle

[mpsampat]

Hi All!

i am trying to convert whole body ct segmentation bundle into a monai deploy MAP.
I dont want to use the MONAIBundleInferenceOperator since
we have a very strong preference for monai deploy MAP over bundles.

I noticed the pre-processing transforms in train.json and inference.json for whole body ct segmentation bundle do not match.
Inference.json has both:
{
"target": "NormalizeIntensityd",
"keys": "image",
"nonzero": true
},
{
"target": "ScaleIntensityd",
"keys": "image",
"minv": -1.0,
"maxv": 1.0
}
while train.json has NormalizeIntensityd but not ScaleIntensityd
{
"target": "NormalizeIntensityd",
"keys": "image",
"nonzero": true
},
Could someone please check it and let me know ?
I can then create an operator for whole body ct segmentation and submit it to monai deploy app github

[yiheng-wang-nv]

HI @tangy5 , could you please help to clarify the difference here?
cc @KumoLiu @wyli @Nic-Ma

[KumoLiu]

Hi @mpsampat, it seems there also have a ScaleIntensityd in the "inference.json".

model-zoo/models/wholeBody_ct_segmentation/configs/inference.json

Line 70 in 99b8be7

"_target_": "ScaleIntensityd",

I hope it helps, thanks!

[mpsampat]

thanks all
I was able to use the ConfigParser and confirm it is there in the pre-processing steps. the code is below:
parser = ConfigParser()
parser.read_config('configs/inference.json')
preprocessing = parser.get_parsed_content("preprocessing")

now we can check the list of transforms as follows:

preprocessing.transforms
(<monai.transforms.io.dictionary.LoadImaged at 0x7f7d5a84cbe0>,
<monai.transforms.utility.dictionary.EnsureTyped at 0x7f7d5ae43f40>,
<monai.transforms.utility.dictionary.EnsureChannelFirstd at 0x7f7d5ae43fd0>,
<monai.transforms.spatial.dictionary.Orientationd at 0x7f7d5ae43430>,
<monai.transforms.spatial.dictionary.Spacingd at 0x7f7d5ae43d30>,
<monai.transforms.intensity.dictionary.NormalizeIntensityd at 0x7f7d5ae43820>,
<monai.transforms.intensity.dictionary.ScaleIntensityd at 0x7f7d5a35efa0>)