Question 1
For question 1, we reproduce the result in Neural style transfer with eager execution and Keras. The code uses neural style transfer techniques to transmute an original picture into the ‘style’ of a second picture.
We tuned the image_shape parameter by changing it from c(128,128,3) to c(256,256,3) and c(512,512,3). This means we broke the pictures in to more pixels which can make a higher resolution of the output picture.
Please see the Rmd file for the whole report.
Question 2
For question 2, we develop an app “Foodie” such that a user takes a photo of a meal and the app will recognize and record the type of food.
Here is our prototype app demo
Tech stack:
- TensorFlow v1.10
- Xcode 10.1
- Food101 Core ML model
- Inception v3 Core ML model
Author
Yun Han, Xinrui Zhang, Juehao Hu, Diyang Wu, Edward Yu
