DeepFaceDrawing: When Sketches Come to Life
Imagine: you draw a face on a napkin — just a few casual lines. And a moment later, a photorealistic portrait appears before you. Science fiction? Not at all — this is DeepFaceDrawing, a project that makes the magic of generative models accessible to everyone.
What's behind the name
DeepFaceDrawing is a neural network model capable of converting simple face sketches into detailed images. Unlike similar solutions, it doesn't require professional drawings: schematic lines are enough, and the system completes them on its own.
The project was developed by researchers and published in ACM Transactions on Graphics. Under the hood — a combination of several neural networks working on a "local-to-global" principle.
Who will find it useful
- Designers for quick idea visualization
- Game developers creating characters
- Artists as a concept art tool
- Hobbyists without artistic training
Key features
-
Works with rough sketches The system correctly interprets even casual sketches, supplementing missing details.
-
Step-by-step processing First, individual face parts are analyzed (eyes, nose, mouth), then a cohesive image is created.
-
Flexible adjustment You can modify already generated images by correcting individual elements.
-
GUI interface The built-in editor allows you to draw directly in the application.
Technical side
The project uses the Jittor framework — a Chinese alternative to PyTorch. You'll need:
- Ubuntu 16.04+
- NVIDIA GPU with CUDA 9.2
- Python 3.7
Installation looks like this:
sudo apt install python3.7-dev libomp-dev
sudo python3.7 -m pip install git+https://github.com/Jittor/jittor.git
python3.7 -m jittor.test.test_example
sh install.sh
How to try it
- Download pre-trained models
- Place them in the 'Param' folder
- Run the demo:
python3.7 demo.py
Where to apply
- Prototyping characters for games and animation
- Education — studying facial anatomy
- Forensics — creating composite sketches
- Art projects — experimenting with visualization
Conclusion: is it worth trying?
DeepFaceDrawing is one of the most impressive tools in its class. If you need to quickly turn ideas into visual images, this project will save hours of work. The main advantage is that it really works with amateur sketches, making it accessible to a wide range of users.
Developers will also find the Jittor-based approach interesting — a great way to get acquainted with this platform. And for those who simply want to experiment with image generation, the project offers a simple graphical interface.
Give it a try — perhaps your sketch will become the next demonstration of neural network capabilities.
Related projects