sage-smoke-detection

Instructions

Step 1: run trainning jupyter notebook and save model

There are two options to train the smoke detection neural network. The first one is to run the jupyter notebook on a Kubernetes cluster (for us it is temporarily going to be Nautilus). The second option is to run it locally assuming that there is a GPU availabe on the local node (the docker image might fail but Tensorflow will not).

Training on a Kubernetes Cluster (Nautilus):

Clone the smoke detection model:

cd training/

Create a persistent volume claim on Nautilus under the Sage namespace (not needed now since it exists):

kubectl create -f training.pvc.yaml

Create a deployment on kubernetes:

kubectl create -f training.deployment.yaml

Attach to a pod and run bash:

kubectl exec -it POD-NAME bash

Run jupyter notebook on pod:

jupyter notebook -—ip=0.0.0.0 -—port=9000

Port forward from pod to local node:

kubectl port-forward POD-NAME 9000:9000

Access the notebook through your desktops browser on http://localhost:9000

Training on a local node(if no kubernetes cluster is available):

If there is no kubernetes cluster available for the user, there is a docker file that can be used to run on a local node (assuming that there is a GPU available).

Build docker image:

docker build --build-arg SAGE_STORE_URL=${SAGE_STORE_URL} --build-arg SAGE_USER_TOKEN=${SAGE_USER_TOKEN} --build-arg BUCKET_ID_MODEL=${BUCKET_ID_MODEL} -t iperezx/training-smokedetect:0.1.0 .

Run docker image:

docker run -it -p 9000:9000 iperezx/training-smokedetect:0.1.0

Attach to container and run jupyter notebook:

docker attach iperezx/training-smokedetect:0.1.0
jupyter notebook --ip 0.0.0.0 --port 9000 --no-browser --allow-root

Access the notebook through your desktops browser on http://localhost:9000

Step 2: Build Docker image for plugin

The docker image is hosted on sagecontinuum. Before building the image make sure that the environment variables (SAGE_STORE_URL,BUCKET_ID_MODEL, HPWREN-FLAG, and TEST_FLAG) are set in the user's local enviroment.

Set enviroment variables:

export SAGE_STORE_URL=https://osn.sagecontinuum.org
export BUCKET_ID_MODEL=BUCKET_ID_MODEL
export HPWREN_FLAG=False
export TEST_FLAG=False

To obtain a token, visit the Sage Authorization UI. The BUCKET_ID_MODEL has been set public so any SAGE user can access the smoke detection models.

Build the image:

docker build --build-arg SAGE_STORE_URL=${SAGE_STORE_URL} --build-arg BUCKET_ID_MODEL=${BUCKET_ID_MODEL} \
--build-arg HPWREN_FLAG=${HPWREN_FLAG} --build-arg TEST_FLAG=${TEST_FLAG} -t sagecontinuum/sage-smoke-detection:0.1.0 .

where the --build-arg adds all the necessary enviroment variables for the Sage Storage API and Sage CLI

Step 3: Run Docker container locally

Example output of the plugin when HPWREN camera API is used( export HPWREN_FLAG=True and export TEST_FLAG=False) :

export SAGE_STORE_URL=https://osn.sagecontinuum.org
export BUCKET_ID_MODEL=BUCKET_ID_MODEL
export HPWREN_FLAG=True
export TEST_FLAG=False

docker build --build-arg SAGE_STORE_URL=${SAGE_STORE_URL} --build-arg SAGE_USER_TOKEN=${SAGE_USER_TOKEN} --build-arg BUCKET_ID_MODEL=${BUCKET_ID_MODEL} --build-arg HPWREN_FLAG=${HPWREN_FLAG} --build-arg TEST_FLAG=${TEST_FLAG} -t sagecontinuum/sage-smoke-detection:0.1.0 .

docker run sagecontinuum/sage-smoke-detection:0.1.0

Starting smoke detection inferencing
Get image from HPWREN Camera
Image url: http://hpwren.ucsd.edu/cameras/L/tje-1-mobo-c.jpg
Description:  Unknown direction Color Original
Perform an inference based on trainned model
Publish
.
.
.
Get image from HPWREN Camera
Image url: http://hpwren.ucsd.edu/cameras/L/tje-1-mobo-c.jpg
Description:  Unknown direction Color Original
Perform an inference based on trainned model
Publish

Example output of the plugin when the pre-recorded MP4 is used( export HPWREN_FLAG=False and export TEST_FLAG=True) :

export SAGE_STORE_URL=https://osn.sagecontinuum.org
export BUCKET_ID_MODEL=BUCKET_ID_MODEL
export HPWREN_FLAG=False
export TEST_FLAG=True

docker build --build-arg SAGE_STORE_URL=${SAGE_STORE_URL} --build-arg SAGE_USER_TOKEN=${SAGE_USER_TOKEN} --build-arg BUCKET_ID_MODEL=${BUCKET_ID_MODEL} --build-arg HPWREN_FLAG=${HPWREN_FLAG} --build-arg TEST_FLAG=${TEST_FLAG} -t sagecontinuum/sage-smoke-detection:0.1.0 .

docker run sagecontinuum/sage-smoke-detection:0.1.0

Starting smoke detection inferencing
Get image from 20190610-Pauma-bh-w-mobo-c.mp4
Image url: 20190610-Pauma-bh-w-mobo-c.mp4
Description: Pre-recorded video
Perform an inference based on trainned model
Publish
.
.
.

Get image from 20190610-Pauma-bh-w-mobo-c.mp4
Image url: 20190610-Pauma-bh-w-mobo-c.mp4
Description: Pre-recorded video
Perform an inference based on trainned model
Publish

Example output of the plugin when the bottom camera on the Wild Sage Node is used( export HPWREN_FLAG=False and export TEST_FLAG=False) :

export SAGE_STORE_URL=https://osn.sagecontinuum.org
export BUCKET_ID_MODEL=BUCKET_ID_MODEL
export HPWREN_FLAG=False
export TEST_FLAG=False

docker build --build-arg SAGE_STORE_URL=${SAGE_STORE_URL} --build-arg SAGE_USER_TOKEN=${SAGE_USER_TOKEN} --build-arg BUCKET_ID_MODEL=${BUCKET_ID_MODEL} --build-arg HPWREN_FLAG=${HPWREN_FLAG} --build-arg TEST_FLAG=${TEST_FLAG} -t sagecontinuum/sage-smoke-detection:0.1.0 .

docker run sagecontinuum/sage-smoke-detection:0.1.0

Coming Soon