Generating Anime Faces via GAN Model, machine learning,deep learning

2019-2020 tavasz

Nincs megadva

Téma leírása

Machine Learning is a cutting edge field in modern-day computer science. It is a fairly new field
with huge strides being made by talented researchers in an effort to create more intelligent
software systems. GANs are basically two-component systems with each component acting as
a network. One component is the generator network which as the name implies tries to create
images based on images that it has seen before in the training data set. The other component is
the discriminator which tries to learn a dataset and then tries to determine whether or not the
image produced by the generator is a real or a fake. GAN stands for Generative Adversarial
Network and as the name implies there is a rivalry of sorts between the two networks. One of
them tries to trick the other while the other tries to not be tricked. As we train for more and more
rounds each of the networks gets better at what it does and in the end, we are hopefully able to
create believable images.
In this project, we will be looking at multiple types of GANs and comparing them before choosing
the most appropriate one for the task at hand. I will also be doing extensive dataset analysis to
determine the optimal dataset size and data parameters to achieve the best results and also
any tactics that are used in the field to optimize how well our model learns from the data
Tasks to be performed by the student will include:
 Learn about different types of GANs and their applications
 Compare different GANs and pick one to use
 Study and analyze similar works in the field
 Determine the appropriate dataset size and relevant specification required to train
 Design a set of parameters to optimize the learning process
 Train using the original Dataset
 Analyze whether or not the results can be improved
 Look into techniques to improve results
 Verify whether or not the project was a success





  • gépi tanulás, machine learning, deep learning, GAN,

Maximális létszám: 3 fő