The following will teach you how to deploy a Human Detector APP on Panorama and monitor the results using Panorama MLOps
Click on Stored Application on the left menu, the below table will have one default app which is ppe_panorama_app.zip
If you wish to upload other application, just click on New Application
Choose uri and then put the bucket uri where you put your zip file of application at
Remember to zip the folder of application
Click on Packaged Application on the left menu, the below table will be empty when you first deploy the PPE
Click on New Package Here you have two option
- If you just want to use default model( coco dataset + yolov5 model ) + sample app then you can just select Packaging in Application Method, you can then skip input a s3 bucket and choose our sample application then click on Submit
- If you want to use your own bucket as the training dataset, then you can choose Training + Packaging and input the S3 Bucket with the training dataset you want, Since we use yolov5 as our model, you will need to prepare a bucket with
- image / your_images.jpg
- label / your_images_label.txt
then again choose the sample application and click on Submit
After that you will see a field with training icon in your application, it means the training begin, you can do other steps and come back after the training finished
The training has two phase, the first phase is Training ( If you only choose Packaging then this step will be skipped )
The next phase is Packaging which will output a graph.json in your app-graph bucket which you could use to deploy Panorama,. When you see ModelStorage has s3://app-graph..., and your status is Completed then it means you are done with the training.
You will need to copy the ModelStorage s3://app-graph... in the Deploy with the PPE step
First click on the left menu, and there’s a Camera Config at the bottom
Click on the New Camera Config on the right side
Input the correspond information for your camera, and click on Submit
After that you can see a camera created in your Camera Config, if you go to Panorama Console, you can see your camera being created under the Data sources
You will first need to create a Panorama device in Panorama Console, after that you will see your device running in the console
Then you will see the device you provisioned show up at your Device Config
After you finish Add a Camera to your PPE, Start training a sample application for your Panorama, and Add a Panorama device to PPE, you can go on with Deploy with PPE
Click on Deployed Application, then click on New Deployment on the right side
Choose the correspond Device ID and Camera ID you just created, and pass the s3 url you get from Start training a sample application for your Panorama and you can randomly name your APP Name
After complete all the field, you can click on Submit
After that, you can see your application status at Deployed Application or you can go to Panorama Console, and click on Deployed applications, you will see your APP is being deployed, please wait for 30-60 minutes, then see your Application is running with green light, you can go to next step
After you Application is running with green light, you can go in front of your camera for 5 seconds , then it will start the detection and sent the data to your database
Click on Alert List
You can see the detection is being return as series of data and image
So you can Monitor if there’s any alert activity or people not wearing mask or people without helmet depends on how you write your APP, in this case, it will be when there’s human walking by.
Click on Alert Verify
You can Manually verify if your model is actually being accurate about the activity at this page
If you think it’s not correct, you can Correct it by redefining the box it captured and Confirm the data
Click on Confirm Data
Then refresh the page, you can see your data successfully modified
You can retrain your data after you have manually confirmed them, so you can make your model to be more precise in the future
Click on Alert Export
Go to your S3 Console and choose the bucket with s3://export-event-xxxxx ( or you can choose your own bucket
Copy the S3 Url to the following field, then Submit
Then you can see your dataset inside your bucket, and use the dataset to Retrain your Model
