In this post i will Show you what you can do with the AI called Custom Vision and Microsoft flow. The original idea comes from two of the Microsoft Program Managers for Microsoft Flow:
- Kent Weare
- Jon Levesque
They made a very good video called „Hot Dog or No Hot Dog“. You can watch it on YouTube:
After watching this video I was very interesting in knowing how this works. So go on to the journey. In my solution i‘m not searching for a Hot Dog but i‘m looking on my work bag which is called „my office because all the belonging I need to work are in there.
For this I take pictures on some things inside of my bag and uploaded them to the Custom Vision Website after I have created there a Project. During the uploading I have to tag my pictures. All of these pictures get two tags: myoffice and work.
Finished the upload I have to train the Custom Vision.
Now we can build the flow to take pictures and send it to the Custom Vision API.
Here you can see an overview of the complete Flow.
The first step is to create a trigger to take a picture or upload a file. For this I’ll use the trigger Button „manually trigger a flow and insert there an input called file.
The next step is to set two variables to set some conditions later in the flow.
The first variable is defined as an Boolean to get true/false and the second variable is an float to get the „Trefferwahtscheinlichkeit“
Now we have to connect to the vision api.
The next action in flow is to predict tags from image. In this action you set an image content the file content from the trigger button Andrej Project ID. For this you have to copy your project id from the Custom Vision Project you have created.
The next step is a condition to process only the items with the selected tag for you process. Here f.e. it is „work“
The next step is in the Yes section and there you will set the float variable with the probability from the Custom vision.
Next a condition will „prüfen“ of the probability is above f.e. 80%.
If yes the Boolean variable is set to yes and a notification is send through the flow app direct to me on my device.