PSGNet: Pure Smoke Image Generation With Gradient and Style Learning.

Area of Science:

• Computer Vision
• Computer Graphics
• Artificial Intelligence

Background:

• Realistic and controllable pure smoke generation is crucial for applications like image editing, visual effects, and security data synthesis.
• Existing methods struggle with generating intricate smoke details and regulating diverse smoke styles.

Purpose of the Study:

• To propose a novel Pure Smoke image Generation Network (PSGNet) for realistic and controllable smoke image synthesis.
• To address the limitations of current methods in generating detailed smoke and controlling its style.

Main Methods:

• Introduced the Pure Smoke image Generation Network (PSGNet) utilizing a gradient and style learning approach.
• Employed smoke shape masks for spatial control (location, contour) and a gradient-based learning framework for physical realism and fine-scale detail.
• Implemented a spatially aware style learning strategy for fine-grained control over smoke attributes (density, color, look).

Main Results:

• The proposed PSGNet effectively generates realistic smoke with rich visual details and customizable styles.
• The gradient learning framework captures subtle smoke structures, enhancing physical realism.
• Spatially aware style learning allows for precise manipulation of smoke attributes.

Conclusions:

• The PSGNet significantly outperforms state-of-the-art methods in generating controllable and realistic pure smoke images.
• The combination of gradient and style learning offers a powerful approach for detailed and customizable smoke synthesis.