NVIDIA Clara Train 3.1
3.1

AIAA Server APIs

Hint

Once you have AIAA server running, you can also check detailed schema for every API at: http://127.0.0.1:$LOCAL_PORT/docs. The $LOCAL_PORT is the port you use when launching the container in Installation.

Admin APIs are host controlled. When you start the AIAA server you can use --admin_hosts option to specify from which client hosts you can invoke the following APIs. The default is * which means no restrictions, that is every client can invoke Admin APIs.

Following are the important ones:

aiaa_admin_apis.png


GET Model

You can get the current configuration for a given model by using this command.

Copy
Copied!
            

curl -X GET "http://127.0.0.1:$LOCAL_PORT/admin/model/clara_ct_seg_spleen_amp" \ -H "accept: application/json"


PATCH Model

If you want to update the config only (without updating the model), you can do so by refreshing the configs and reloading an existing model with new configs.

Copy
Copied!
            

curl -X PATCH "http://127.0.0.1:$LOCAL_PORT/admin/model/clara_ct_seg_spleen_amp" \ -H "Content-Type: application/json" \ -d @config_aiaa.json


PUT Model

This API will help to load a model into AIAA server.

Copy
Copied!
            

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

Hint

Refer to Loading Models for detailed usage of loading a model.


DELETE Model

You can remove the model from AIAA server using this API.

Copy
Copied!
            

curl -X DELETE "http://127.0.0.1:$LOCAL_PORT/admin/model/clara_ct_seg_spleen_amp"

Note

If you have multiple AIAA server running and they share common workspace for saving all the models and configurations, you should enable --auto_reload option while starting the AIAA Server. This will help to keep all the AIAA Server in sync when a model is loaded/updated/removed.


APIs that are designed for all regular clients.

models

List models loaded in AIAA Server

Copy
Copied!
            

# Query All Models curl -X GET "http://127.0.0.1:$LOCAL_PORT/v1/models" # Query Matching Models for a given Label and Type curl -X GET "http://127.0.0.1:$LOCAL_PORT/v1/models?label=spleen&type=segmentation" # Query Specific Model curl -X GET "http://127.0.0.1:$LOCAL_PORT/v1/models?model=clara_ct_seg_spleen_amp"


segmentation

Run Segmentation model in AIAA Server for an input image.

Copy
Copied!
            

curl -X POST "http://127.0.0.1:$LOCAL_PORT/v1/segmentation?model=clara_ct_seg_spleen_amp&output=image" \ -H "accept: multipart/form-data" \ -H "Content-Type: multipart/form-data" \ -F "params={}" \ -F "image=@spleen.nii.gz;type=application/x-gzip" \ -o output_image.nii.gz


dextr3d

Run DExtr3D in AIAA Server for an input image and extreme points.

Copy
Copied!
            

curl -X POST "http://127.0.0.1:$LOCAL_PORT/v1/dextr3d?model=clara_ct_annotation_spleen_amp&output=image" \ -H "accept: multipart/form-data" \ -H "Content-Type: multipart/form-data" \ -F "params={\"points\":\"[[93,106,64],[40,108,64],[29,66,64],[47,20,64],[93,32,64],[99,68,64]]\"}" \ -F "image=@cropped_spleen.nii.gz;type=application/x-gzip" \ -o output_image.nii.gz


deepgrow

Run DeepGrow in AIAA server given foreground/background points.

Copy
Copied!
            

curl -X POST "http://127.0.0.1:$LOCAL_PORT/v1/deepgrow?model=clara_deepgrow&output=image" \ -H "accept: multipart/form-data" \ -H "Content-Type: multipart/form-data" \ -F "params={\"foreground\":\"[[155, 189, 78], [184, 192, 78], [114,225,78]]\", \"background\":\"[]\"}" \ -F "image=@spleen.nii.gz;type=application/x-gzip" \ -o output_image.nii.gz


inference

Run inference on any generic model in AIAA server.

Please refer to Model Config for details on type of models supported in AIAA.

Copy
Copied!
            

curl -X POST "http://127.0.0.1:$LOCAL_PORT/v1/inference?model=my_model&output=image" \ -H "accept: multipart/form-data" \ -H "Content-Type: multipart/form-data" \ -F "params={\"mydata\":[1,2,3]}" \ -F "image=@spleen.nii.gz;type=application/x-gzip" \ -o output_image.nii.gz

Note

You can use /inference API to run model pipelines, classification models etc.


mask2polygon

Given a 3D mask in NIFTI file, generate polygons (0 or more) on each 2D slices.

Copy
Copied!
            

curl -X POST "http://127.0.0.1:$LOCAL_PORT/v1/mask2polygon" \ -H "accept: application/json" \ -H "Content-Type: multipart/form-data" \ -F "params={ "more_points": 10 }" \ -F "image=@mask2polygon.input.nii.gz;type=application/gzip"


fixpolygon

Adjust the polygons to a better fit 2D/3D polygons.

Copy
Copied!
            

# 2D Image/Input => Output: Result Json curl -X POST "http://127.0.0.1:$LOCAL_PORT/v1/fixpolygon?output=results" \ -H "accept: multipart/form-data" \ -H "Content-Type: multipart/form-data" \ -F "params=`cat fixpolygon.input.json`" \ -F "image=@fixpolygon.input.png;type=image/png" \ -o new_polygons.json # 2D Image/Input => Output: Image curl -X POST "http://127.0.0.1:$LOCAL_PORT/v1/fixpolygon?output=image" \ -H "accept: multipart/form-data" \ -H "Content-Type: multipart/form-data" \ -F "params=`cat fixpolygon.input.json`" \ -F "image=@fixpolygon.input.png;type=image/png" \ -o new_mask.png # 3D Image/Input => Output: Image curl -X POST "http://127.0.0.1:$LOCAL_PORT/v1/fixpolygon?output=image" \ -H "accept: multipart/form-data" \ -H "Content-Type: multipart/form-data" \ -F "params=`cat fixpolygon.3d.input.json`" \ -F "image=@label.nii.gz;type=application/x-gzip" \ -o new_label.nii.gz

The following is a schema for the FixPolygon input json file.

Field

Description

dimension

Dimension to represent 2D or 3D

poly

Represents new polygon points (not required if vertex_offset is used)

prev_poly

Represents current polygon points

slice_index

Represents current slice in case of 3D image

polygon_index

Represents modified polygon index among poly

vertex_index

Represents modified vertex index

vertex_offset

Represents new vertex after adding offset (x,y) for 2D/3D

propagate_neighbor

Propagation size for neighborhood

propagate_neighbor_3d

Propagation size for neighborhood in case of 3D

Clients can choose to create a session for an image to reduce the time in consecutive requests.

This is recommended if users are interacting using DExtr3D or DeepGrow API with 3D volumes. Users pay this one time cost to upload the image to the server, and the following requests will be faster because the image does not need to be uploaded again. You can specify expiry in seconds when creating a session to decide the time that this session would live.

PUT Session

Create a new AIAA session and upload an image as part of the session.

Copy
Copied!
            

curl -X PUT "http://127.0.0.1:$LOCAL_PORT/session/?expiry=30" \ -H "accept: application/json" \ -H "Content-Type: multipart/form-data" \ -F "image=@spleen.nii.gz;type=application/x-gzip"

You can also provided a DICOM server address so that AIAA will fetch images from there.

Let’s assume you have a DICOM server DCM4CHEE running at ip 0.0.0.0, and the AE DCM4CHEE listens on port 11112. Then you put the ip, port, ae_title and id of patient, study and series in a json file (data.json) as follows:

Copy
Copied!
            

{ "dicom": { "server_address": "0.0.0.0", "server_port": 11112, "ae_title": "DCM4CHEE", "query_level": "PATIENT", "patient_id": "ProstateX-0004", "study_uid": "1.3.6.1.4.1.14519.5.2.1.7311.5101.170561193612723093192571245493", "series_uid": "1.3.6.1.4.1.14519.5.2.1.7311.5101.206828891270520544417996275680" } }

Now you can ask AIAA to fetch the data from DICOM server and create a session using the following command:

Copy
Copied!
            

curl -X PUT "http://127.0.0.1:$LOCAL_PORT/session/?expiry=30" \ -H "accept: application/json" \ -d "@data.json"

AIAA also supports getting data using HTTP requests. Assume you have a image server running and you can download the data using a url. You can prepare a data.json file that contains this information. For example:

Copy
Copied!
            

{ "http": { "url": "http://0.0.0.0/data/projects/NV_CLARA/subjects/XNAT_S00001/experiments/XNAT_E00002/scans/49621430/resources/DICOM/files", "params": { "format": "zip" }, "method": "GET" } }

Then you ask AIAA to fetch the data based on this config, that is using HTTP Get method on this url. AIAA will create the session after the data is fetched.

Copy
Copied!
            

curl -X PUT "http://127.0.0.1:$LOCAL_PORT/session/?expiry=30" \ -H "accept: application/json" \ -d "@data.json"

Note

AIAA utilizes pynetdicom to do c_get to fetch the data from DICOM server. You can refer to their documentation for more information.

AIAA uses requests library to handle the HTTP fetch. Please refer to request documentation for optional arguments.


GET Session

Retrieve saved session/image document available from the server.

Copy
Copied!
            

curl -X GET "http://127.0.0.1:$LOCAL_PORT/session/[session_id here]?image=false" \ -H "accept: application/octet-stream"


DELETE Session

Close an existing session.

Copy
Copied!
            

curl -X DELETE "http://127.0.0.1:$LOCAL_PORT/session/[session_id here]"


© Copyright 2020, NVIDIA. Last updated on Feb 2, 2023.