Related Experiment Video
Updated: Jul 26, 2026
07:05
Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
Published on: June 18, 2021
2.8K
AttentionPainter: An Efficient and Adaptive Stroke Predictor for Scene Painting
IEEE Transactions on Visualization and Computer Graphics
|October 6, 2025
Summary
AttentionPainter significantly speeds up neural painting by predicting all strokes in one step, unlike older methods. This efficient approach enhances training and improves image reconstruction quality for digital art applications.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Digital Art
Background:
- Stroke-based Rendering (SBR) decomposes images into strokes for painting generation.
- Existing Neural Painting methods face challenges with slow inference and unstable training.
Purpose of the Study:
- To develop an efficient and adaptive single-step neural painting model.
- To improve inference speed and training stability in stroke-based rendering.
Main Methods:
- Introduced AttentionPainter, a novel scalable stroke predictor for single-forward-pass parameter prediction.
- Developed a Fast Stroke Stacking algorithm for accelerated training (13x).
- Implemented Stroke-density Loss to enhance detail reconstruction using smaller strokes.
Main Results:
- AttentionPainter achieves faster inference than previous neural painting techniques.
- The Fast Stroke Stacking algorithm significantly boosts training efficiency.
- Stroke-density Loss improves the quality of reconstructed images.
- AttentionPainter outperforms state-of-the-art neural painting methods in experiments.
Conclusions:
- AttentionPainter offers an efficient and effective solution for single-step neural painting.
- The model's speed and improved reconstruction quality benefit digital art creation.
- A Stroke Diffusion Model application demonstrates potential for inpainting and editing.
