GTC2020 Workshop TrainingData.io Repo

This repos contains sample project for AI-Assited Labeling for Radiology AI using NVIDIA Clara on TrainingData.io

How to use this Docker image: Quickstart Guide for TrainingDataio/tdviewer

1. Use Dockerfile to build a docker image

docker build -t gtc2020-trainingdataio .

2 Run a container from this image

docker run -p 8888:8888 9090:9090 8090:8090 8000:8000 --runtime=nvidia gtc2020-trainingdataio /bin/bash

Once the container is running,

cd /workspace/content

3. Login to hub.docker.com:

docker login hub.docker.com

4. Login to nvcr.io

docker login hub.docker.com

5. Create a directory on your disk to store TD.io database. For example "/home/user/db"

mkdir -p  /path/to/db/directory

6. Create a directory to place images and videos (dataset assets). For example: "/home/user/images"

mkdir -p /path/to/images/directory

7. Make sure /tmp directory exists and is readable & writable

mkdir -p /tmp

8. Run Docker image providing mount point for database-folder and mount point for images-folder.

export DB_MOUNT=/path/to/temp/directory && export IMAGE_MOUNT=/path/to/image/directory && docker-compose -f docker-compose.ngx.yml up -d

9. Load model in Clara:

curl -X PUT "http://0.0.0.0:5000/admin/model/segmentation_ct_liver_and_tumor" -H "accept: application/json" -H "Content-Type: application/json" -d '{"path":"nvidia/med/segmentation_ct_liver_and_tumor","version":"1"}'

10. Open Webbrower: https://app.trainingdata.io/v1/td/login

Login with oAuth or username & password

11. Start AI-Assisted Labeling

  On tab “Labeling Jobs” select “On-Premises Labeling Job”
  Click “Start Labeling”
  Observe http://127.0.0.1 loads in web-browser

12. Optional (3D Slicer):

• Download 3D Slicer: https://trainingdataio.s3.amazonaws.com/SlicerLatest-TDIO-GTC2020Workshop.tgz
• Extract and open Slicer executable
• Setup /workspace/content/slicer-plugin as module in Slicer
• Open the labeling project in "Open in 3D Slicer"

Additional Information

Manage On-Premises TrainingData Labeling

How to create on-premises datasets?

How to create labeling instructions?

How to create labeling jobs?

How to distribute labeling jobs among annotators and reviewers?

Supported export formats for annotated data?

Technical Support

Email: [email protected]