Time to work !

Today we will actually be getting started on the project. We will have three intense weeks to works exclusively on this.

Our plan is :

• Fisrt week : familiarize with the technologies we will be using : we chose to work with the Tensorflow API for Python. Then we will start to work on our Neural Network.
• Second week : we hope we will be able to start training our network and maybe have first results. Besides we will start to build the website to host the application.
• Thrid week : if we are in time we will use this week to experiment with our program and try to have satisfing results. And we will finish the website.

About the technology

Training the network :

For this project we built a GAN (Generative Adversarial Network) using the Tensorflow API with Python.

This is what our GAN looks like :

As you can see, our network works with different inputs. We pass it an image of random noise and a real image from our database but randomly masked. The Generator will then try to complete the image where the mask was applied. Then it passes its result to the Discriminator and tries to fool it, making it believe that his image is an image from the database. How much the Discriminator was fooled is indicated by a precision, basically indicating if the Generator did a good job. Knowing this the Generator will try again and so on until the end of the training.

The Generator is built like this :

And the Discriminator like this :

Using the network :

Once we are satisfied with the results of the training of our network, we can use it. But how ? Basically we use Tensorflow to transform our Python model in a Javascript model so that we can use it in a website.