Unable to upload model from NGC for Clara Train v2.0

Tried to upload the annotation_mri_hippocampus model, both using local download and then upload to the server or direct push from NGC server. Both fails with:

500 Internal Server Error

Internal Server Error

The server encountered an internal error and was unable to complete your request. Either the server is overloaded or there is an error in the application.

No output in the console running the server.

Have made it work earlier but then for older version of Clara Train.

Apparently the specific model is not supported by version 2.0.
Please note that the v2.0 container is compatible only with the below list of models, v1.x models are not compatible with v2.0 container.
clara_ct_seg_spleen_amp
clara_ct_seg_spleen_no_amp
clara_ct_seg_liver_and_tumor_amp
clara_ct_seg_liver_and_tumor_no_amp
clara_mri_seg_brain_tumors_br16_full_amp
clara_mri_seg_brain_tumors_br16_full_no_amp
clara_xray_classification_chest_amp
clara_xray_classification_chest_no_amp
clara_mri_annotation_brain_tumors_t1ce_tc_amp
clara_mri_annotation_brain_tumors_t1ce_tc_no_amp
clara_ct_annotation_spleen_amp
clara_ct_annotation_spleen_no_amp
clara_mri_fed_learning_seg_brain_tumors_br16_t1c2tc_amp
clara_mri_fed_learning_seg_brain_tumors_br16_t1c2tc_no_amp
clara_mri_seg_brain_tumors_br16_t1c2tc_amp
clara_mri_seg_brain_tumors_br16_t1c2tc_no_amp

Any suggestions on how to access an older version of clara train SDK?

Hello Daniel,

You should be able to pull older tags from NGC:

https://ngc.nvidia.com/catalog/containers/nvidia:clara-train-sdk/tags

That being said; you can also try to bring your own model (BYOM) to V2; more information here:

https://docs.nvidia.com/clara/tlt-mi/clara-train-sdk-v2.0/nvmidl/byom.html

Thanks.

how do you interact with your AIAA server?

are u using MITK/Slicer

or just curl command?

Managed to pull an older tag. Had a bit of an issue at first cause the naming of the older tag was slightly different.

For interaction with the AIAA server I use MITK Workbench.

Got it,
if you want to run that model in new version of AIAA
you can also use the following config.json:

{
  "version": "2",
  "type": "annotation",
  "labels": [
    "hippocampus"
  ],
  "description": "A pre-trained model for volumetric (3D) annotation of the hippocampus head and body from mono-modal MRI image",
  "pre_transforms": [
    {
      "name": "LoadNifti",
      "args": {
        "fields": "image"
      }
    },
    {
      "name": "AddLabelPoints",
      "args": {
        "image_field": "image",
        "label_field": "label",
        "points": "points"
      }
    },
    {
      "name": "ConvertToChannelsFirst",
      "args": {
        "fields": [
          "image",
          "label"
        ]
      }
    },
    {
      "name": "NormalizeNonzeroIntensities",
      "args": {
        "fields": "image"
      }
    },
    {
      "name": "AddExtremePointsChannel",
      "args": {
        "image_field": "image",
        "label_field": "label"
      }
    }
  ],
  "inference": {
    "image": "image",
    "scanning_window": false,
    "batch_size": 1,
    "roi": [
      128,
      128,
      128
    ],
    "padding": 20.0,
    "tf": {
      "input_nodes": {
        "image": "NV_MODEL_INPUT"
      },
      "output_nodes": {
        "model": "NV_MODEL_OUTPUT"
      }
    },
    "trtis": {
      "input_channels": {
        "image": 2
      },
      "output_channels": {
        "model": 3
      },
      "gpu_instance_count": 1,
      "max_batch_size": 8
    }
  },
  "post_transforms": [
    {
      "name": "ArgmaxAcrossChannels",
      "args": {
        "fields": "model"
      }
    },
    {
      "name": "CopyProperties",
      "args": {
        "fields": [
          "model"
        ],
        "from_field": "image",
        "properties": [
          "affine"
        ]
      }
    }
  ],
  "writer": {
    "name": "WriteNifti",
    "file_ext": "",
    "args": {
      "field": "model",
      "dtype": "uint8"
    }
  }
}

Let me know if this does not work