Infer lung tumor by Nvidia's pre-trained model [Deploy SDK]

Hi,

I have lung’s CT volume data, which has tumor.
I want to use Nvidia Deploy SDK and pre-trained model, “segmentation_ct_lung_tumor”.
( https://ngc.nvidia.com/catalog/models/nvidia:med:segmentation_ct_lung_tumor )

Then, I have some questions.

(1) Can “segmentation_ct_lung_tumor” model infer where lung tumor on CT images ?

(2) Can I use “segmentation_ct_lung_tumor” model directly on Nvidia Deploy SDK ?
I mean I will use the model without Annotation Workflow or Transfer Learning and Training Workflow on Nvidia Train SDK.
Need I do “Annotation”, “Transfer” and “Training” steps ?

Hi yun_univ,

(1) Can “segmentation_ct_lung_tumor” model infer where lung tumor on CT images ?

Correct. In the inference out, 0 means background and 1 means lung tumor.

(2) Can I use “segmentation_ct_lung_tumor” model directly on Nvidia Deploy SDK ?

Yes, the bundled Operator/Pipeline for Lung Tumor (in R4[0.4.0]) is using the pre-trained model from https://ngc.nvidia.com/catalog/models/nvidia:med:segmentation_ct_lung_tumor, and model file (models/model.trt.pb) was copied/renamed to ‘model.graphdef’ with ‘config.pbtxt’ file to use NVIDIA inference server(https://docs.nvidia.com/deeplearning/sdk/tensorrt-inference-server-guide/docs/model_repository.html)

You don’t need “Annotation”, “Transfer” and “Training” steps to use it.

<b>root@clara:/clara/common/models/segmentation_ct_lung_tumor_v1#</b> tree .
.
|-- 1
|   `-- model.graphdef
`-- config.pbtxt

Please see the following document:
https://docs.nvidia.com/clara/deploy/sdk/Applications/Pipelines/LungTumorPipeline/public/docs/README.html
https://docs.nvidia.com/clara/deploy/sdk/Applications/Operators/ai/app_lungtumor/public/docs/README.html

And, you can download/install the latest version (0.4.1, 12/10/2019) from the following links:

https://ngc.nvidia.com/catalog/model-scripts/nvidia:clara_deploy_sdk/setup
https://docs.nvidia.com/clara/deploy/ClaraInstallation.html

Hi, gigony.

Thank you for your replying and guiding.
I will try it.