An Example: Implement 2D colonoscopy segmentation model into MONAI Label

Environment

The app was developed with the official MONAI docker container (https://hub.docker.com/r/projectmonai/monai, v0.8.1) and MONAI Label v0.4.dev2213.
MONAI Label installation:

pip install monai==0.8.1
pip install monailabel-weekly==v0.4.dev2213

Pre-trained model

Please put the pre-trained models of PraNet-19 and Res2Net weights into the folder colonoscopy_app/model/.
The pre-trained models can be downloaded from the PraNet repo: https://github.com/DengPingFan/PraNet#31-trainingtesting

Sample data

From Kvasir-SEG dataset https://datasets.simula.no/kvasir-seg/

Note:

• Replace monailabel transforms.py (located in /your/pythonlib/path/of/monailabel/deepedit/multilabel/transforms.py) with the provided transforms.py
• Replace monai dice.py (located in /your/pythonlib/path/of/monai/losses/dice.py) with the provided dice.py

The full repo with pre-trained models can be downloaded here: https://drive.google.com/file/d/1tetR6TtWcsp_g6mkHmsk8kFvDuDpiJdA/view?usp=sharing

Steps to Build Monai on Clara AGX Dev Kit

The pre-build docker image for PyTorch/TorchVision/Monai/MonaiLabel is provided on docker hub.

docker pull eahung/cagx_monai

SW version in the image

• Ubuntu 20.04
• CUDA 11.2.0
• CUDNN 8
• PyTorch 1.8.0
• TorchVision 0.9
• TorchAudio 0.8
• Monai 0.8.1

Steps to build the image (It takes several hours and a large amount of disk space to build the image)

1. Pull docker image nvidia/cuda:11.2.0-cudnn8-devel-ubuntu20.04

2. Build PyTorch and TorchVision

   • Suggested read: https://qengineering.eu/install-pytorch-on-jetson-nano.html

3. Build Monai (https://docs.monai.io/en/stable/installation.html)

BUILD_MONAI=1 
pip install --no-build-isolation git+https://github.com/Project-MONAI/MONAI#egg=monai

4. Build Monai Label (https://docs.monai.io/en/stable/installation.html)

pip install git+https://github.com/Project-MONAI/MONAILabel#egg=monailabel

Contributors

---

Eddie Huang, NVIDIA, [email protected]
Eason Hung, NVIDIA, [email protected]