With the rapid development of AI technology, especially generative large models, AI has penetrated into various fields, including PCs and operating systems. The deepin operating system is working hard to deeply integrate with AI to allow users to enjoy a smarter and more efficient experience.
Previously, deepin selected three major applications commonly used by users: intelligent global search, email, and browser, and took the lead in enabling AI capabilities. Recently, Deepin Image has quietly updated its AI image retouching function, allowing us to take a closer look at the changes that have taken place, from installation to practical operation to the technical implementation behind it!
Easy installation
After updating the system to deepin V20.9, you can install the "Picture Viewing AI Plug-in" in the app store with one click:
Of course, you can also install it directly through the command:
sudo apt install imageenhanceRich localized image AI functions
In the latest picture viewing AI plug-in, 8 AI picture editing functions have been added in one go. It does not rely on the Internet online model, nor is it restricted by hardware constraints of computing power. It can be easily run on a single PC.
Image colorization: automatically add color to black and white images or monochrome images;
Improve resolution: Increase the pixel density of the image to make the image details clearer and more concrete;
Blur background: Reduce the clarity of the image background to make the subject more prominent;
Delete background: Remove the image background to make the subject independent or replace it with a new background ;
Hand-drawn comics: Convert real images into hand-drawn comic style;
2D comics: Convert real images into 2D comic style;
3D comics: Convert real images into 3D comic style;
Sketch: Convert real images into sketch style.
Let’s see it in action:
Picture coloring
Increase resolution
blurred background
Remove background
hand drawn comics
2D comics
3D comics
sketch
Exploring the technical implementation behind
Compared with large models that often require powerful computing power and hardware equipment, what are the technologies behind localized AI models?
Style conversion: Using the new domain-calibrated image translation model DCT-Net (Domain-Calibrated Translation), using a small sample of style data, a high-fidelity, robust, and easy-to-expand portrait style conversion model can be obtained, and through end-to-end Inference quickly obtains style transfer results;
Image resolution improvement: Real-ESRGAN proposes to simulate real complex degradation through multiple degradation methods. Compared with the previous simple down-sampling, it can better handle real low-resolution scenes;
Image coloring: Using the latest SOTA image coloring algorithm , it can generate natural and vivid color results for input black and white images.
Easter egg: deploy other AI frameworks on deepin
The AI capabilities of the deepin operating system are so strong, can it be deployed and used by other mainstream frameworks? The answer is yes, let’s challenge OpenVINO first.
OpenVINO is a comprehensive tool suite from Intel for rapid deployment of vision applications and solutions. It supports more than 150 CNN network structures for computer vision. It is a Pipeline toolset that can complete various capabilities required for algorithm model deployment.
Environment installation
Prerequisites: Conda, Git.
Installation command:
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple -r requirement.txtrequirements.txt :
openvino==2023.1
openvino-dev==2023.1
git+https://github.com/huggingface/optimum-intel.git
onnx==1.14.1Model conversion
After the environment is installed, for the PyTorch model, you need to convert it into an ONNX model, and then use the command to convert it into an IR model:
mo --input_model model.onnxAfter the model is converted, OpenVINO can run the relevant demo normally for image conversion.
This is an AI era full of infinite possibilities. With the support of AI, in-depth picture viewing can make every image a unique expression of our soul, making every sharing a beautiful dialogue with the world. Let us explore the infinite possibilities of visual art together and make life better!
Content source: deepin community