4. What are we trying to achieve?
• Transform any hands-drawn design into a HTML code
5. How can we build this?
Split up the problem
Handwriting recognition
Recognize UI control
Interpret positioning
Can we code this?
How to interpret graphical
input?
Many variations in
handwriting
Edge detection, pattern
matching, if-then-else
Can we use AI?
DIY model
Choose existing models
Sample labeled data
Model training
6. AI…OK, give it to me!
https://towardsdatascience.com/r-cnn-fast-r-cnn-faster-r-cnn-yolo-object-detection-algorithms-36d53571365ehttps://towardsdatascience.com/handwriting-recognition-using-tensorflow-and-keras-819b36148fe5
http://www.fki.inf.unibe.ch/databases/iam-handwriting-database
8. AI…I’m not OK…
• How do I know about these models?
• What’s a neural network anyway?
• Where do I get the data to train these models?
• How do I integrate these models into my app?
12. Vision Service
• Text and handwriting detection
• Text and position
• Off the shelve – requires no
training
• Asynchronous API
13. Vision Service – Getting started
1. Create an account
• Pricing tier
• Region
2. Get the subscription key
3. Set HTTP headers
• Ocp-Apim-Subscription-Key
• Content-Type
4. Invoke method
• RecognizeText
• GetTextOperationResult
15. Custom Vision Service
• Object detection on your data
• Objects and position
• Bring your own data – limited dataset
due to transfer learning
• Asynchronous API
• Training
• Prediction
16. Custom Vision Service – Getting started
1. Create an account
2. Upload and tag images
3. Train Model
4. Get the training & prediction
keys
5. Set HTTP headers
• Training-Key | Prediction-Key
6. Invoke method
• TrainProject
• GetIteration
• DetectImage
19. Recap
• AI can help you turn data into programs
• AI is not only for Data Scientists
• Bring your own Model
• Bring your own Data
• Bring your own HTTP request
• Get started at https://aka.ms/jsconfbe/cognitivedocs
20. “I choose a lazy person to do
a hard job. Because a lazy
person will find an easy way
to do it.”