How Can a Neural Network Transform Faces into Alien Art?

  • Thread starter Thread starter btb4198
  • Start date Start date
  • Tags Tags
    Network Neural
Click For Summary

Discussion Overview

The discussion revolves around the creation of a neural network capable of transforming human faces into alien art, specifically through the use of deep learning techniques. Participants explore the requirements for training data, the types of neural networks suitable for this task, and existing technologies that may assist in achieving the desired outcome.

Discussion Character

  • Exploratory
  • Technical explanation
  • Debate/contested

Main Points Raised

  • One participant suggests that a neural network could learn to transform faces into alien art by training on a large dataset of original and corresponding alien images, emphasizing the importance of having sufficient training data.
  • Another participant mentions they are building a database of face images that will be redrawn by an artist, questioning what type of neural network would be appropriate for their goal, as they are not focused on object detection.
  • A participant provides a link to a resource that details neural network training, referencing a large dataset of handwritten letters as an example, while expressing skepticism about the feasibility of training with only a few dozen images.
  • There are inquiries about how existing applications like Prequel achieve their cartoon transformations, with some participants doubting the necessity of a large dataset for such tasks.
  • One participant suggests using existing facial recognition libraries as a starting point to avoid unnecessary complexity in development.
  • A recommendation is made for using Generative Adversarial Networks (GANs) or Cycle GANs for the image transformation task, particularly for handling unpaired datasets.

Areas of Agreement / Disagreement

Participants express varying opinions on the amount of training data required, the specific neural network architectures to use, and the methods employed by existing applications. No consensus is reached on these points, indicating multiple competing views remain.

Contextual Notes

Participants mention the need for a substantial amount of training data, but there is uncertainty about the exact quantity necessary for effective training. The discussion also highlights the potential for different approaches to achieve similar outcomes, reflecting the complexity of the topic.

Who May Find This Useful

This discussion may be of interest to individuals exploring neural networks, deep learning applications in art, and those curious about image transformation techniques in machine learning.

btb4198
Messages
570
Reaction score
10
I have a picture of a guy's face and I had an artist draw him as an alien. How do I make a neural network that would take in the original picture and learn how to draw the picture the that artist make?

I want to make a filter, that if you can enter in any guy's face, and the program would draw is like if it was the artist.
 
Technology news on Phys.org
btb4198 said:
I have a picture of a guy's face and I had an artist draw him as an alien. How do I make a neural network that would take in the original picture and learn how to draw the picture the that artist make?

I want to make a filter, that if you can enter in any guy's face, and the program would draw is like if it was the artist.
That's interesting and novel. Deep learning neural networks learn from large numbers of examples. If you have 100,000 example pictures of people plus 100,000 examples of how the corresponding alien pictures would look, then you can use those as training data to train a neural network.

The difficulty is not in the deep learning, but the collections of the training data.
 
anorlunda,

I have a lot of pictures of different faces and I am going to have the artist do the same thing to them.
So I am working on building a big database.
but again what kind of neural network should I make? All the videos I have seen on the subject, seem to be about finding an object in an image.

but I am not trying to find something
 
http://neuralnetworksanddeeplearning.com/

That link it to a free online book. The book explains in detail how to do what you want. It illustrates using a NIST data base of more than 250,000 images of handwritten letters identified by real people.
btb4198 said:
I have a lot of pictures
I hope you really mean that. A quarter million really is a lot. If you mean a few dozen, it won't work well.
 
Question,

How does you think Prequel does it's cartoon me?
Do you think they have a quarter Million of face pictures, because I really did not think it would take that many. That seems excessive.
Right now I have 139 different face Pictures redrawing by the same artist.
 
btb4198 said:
Question,

How does you think Prequel does it's cartoon me?
Do you think they have a quarter Million of face pictures, because I really did not think it would take that many. That seems excessive.
Right now I have 139 different face Pictures redrawing by the same artist.
I don't know about Prequel, but it is possible that they did it by some means other than deep learning trained neural networks.
 
  • Like
Likes   Reactions: pbuk
There are already libraries that can recognize facial features. (see e.g. DeepFace) I would start there. The first thing to learn in programming is to not reinvent the wheel.
 
Generative adversarial network (GAN) is what you're looking for. Or cycle GAN which can be used for unpaired images (e.g. a databased of human pictures and a separate one of aliens).

The topic you can look up is image-to-image translation.
 
Last edited:
Here is a video explaining how it works.

 
  • #10
Jarvis323 said:
Here is a video explaining how it works.


Thanks, I am going to watch his videos on this
 

Similar threads

  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 1 ·
Replies
1
Views
3K
  • · Replies 2 ·
Replies
2
Views
3K
  • · Replies 6 ·
Replies
6
Views
2K
  • · Replies 18 ·
Replies
18
Views
4K
  • · Replies 5 ·
Replies
5
Views
3K
Replies
3
Views
2K
  • · Replies 1 ·
Replies
1
Views
2K
Replies
4
Views
5K