Determine if there is hotdog in the picture.
- Download necessary library listed in app.py (I forgot to create python virtual environment and that will never happen again).
- Change your directory to current project folder.
- run
python3 app.pyorpython app.py
-
Create tensorflow model
- Find training and testing data. I use this.
- Categorize the image into different folders (hotdog and not hotdog) for tensorflow to understand.
- [Optional] Perform image augmentation.
- Check labels and adjust if necessary. (IMPORTANT!)
- [Optional] Create prefetch buffer and callbacks function for run time optimziation.
- Come up with tensorflow models and metrics to determine model performance (I use model accuracy and binary cross entropy).
- Train model with training and validation data.
- Test model with testing data.
- Determine how to improve model (overfitting or underfitting, dropout layers, adjust learning rate, try new activation functions, image augmentation, more data...)
- Save weights (not the model, ONLY WEIGHTS, I have to set model structure again in app.py)
-
Design front end
- Use HTML, CSS, and JavaScript to design web application
-
Do back end
- Use flask to handle image processing, decision machine and output result
Harder than I expected but it was fun.