Replacing a Spleen model with a heart model in Clara AIAA

Hi Brad,

Hope you are well and your family is safe. We have been using Clara for imaging segmentation problems. We have worked through the following tutorial → NVIDIA NGC

and it works great using the Clara cloud deployment on AWS → Accelerated model training and AI assisted annotation of medical images with the NVIDIA Clara Train application development framework on AWS | Containers

However, if we switch the spleen model with let’s say the heart model → HeaortaNet_clara_ct_seg_heart_and_aorta → the following step →

curl http://clara-LoadB-ABLH1BWGGPQ8-799966118.us-west-2.elb.amazonaws.com/v1/models

in the AWS blog still reports with the spleen model. What are we doing incorrectly?

Hi
Thanks for your interest in Clara Train SDK. Glad you managed to get it working on AWS.
The artical you point to is a bit old (over a year old). It is using clra train V3.
Since then we have release clara train V4.0 last April based on MONAI which uses PyTorch.

I am not sure how you switched the model? you need to copy the mmar structure change the model file, adjust the aiaa config files then do the curl command to load the model into aiaa. You could see examples on github repo at the depricated folder

curl -X PUT "http://127.0.0.1/admin/model/clara_ct_seg_spleen_amp" \
     -F "config=@clara_ct_seg_spleen_amp_v1/config/config_aiaa.json;type=application/json" \
     -F "data=@clara_ct_seg_spleen_amp_v1/models/model.trt.pb"

We strongly encourage you to move to our recent release as the AIAA has a cleaner command line interface instead of doing curl commands
Please check out the notebooks to get you started clara-train-examples/PyTorch/NoteBooks at master · NVIDIA/clara-train-examples · GitHub
Please note, since then we switched all our models on NGC to be Pytorch based instead of TF. Note that AIAA can still server both TF an PT models. However the configuration files has changed.

Hope this helps