Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Emergence of Painting Ability via Recognition-Driven Evolution

Yi Lin, Lin Gu, Ziteng Cui, Shenghan Su, Yumo Hao, Yingtao Tian, Tatsuya Harada, Jianfei Yang

TL;DR

The paper asks whether painting ability can emerge under evolutionary pressures that optimize visual communication. It introduces a two-branch system that generates vector paintings from images: a Stroke Branch that optimizes Bézier strokes and a Palette Branch that learns a restricted colour palette, both guided by a CLIP-based high-level recognition objective and a scenario complexity estimator. The key contributions include emergent painting ability under evolutionary pressure, a dataset and estimator for painting scenario complexity, and demonstrations of high recognition accuracy with few strokes and colours plus potential for low bit rate image compression. The work connects AI driven painting emergence to historical artistic practices and presents practical implications for efficient content generation and compression.

Abstract

From Paleolithic cave paintings to Impressionism, human painting has evolved to depict increasingly complex and detailed scenes, conveying more nuanced messages. This paper attempts to emerge this artistic capability by simulating the evolutionary pressures that enhance visual communication efficiency. Specifically, we present a model with a stroke branch and a palette branch that together simulate human-like painting. The palette branch learns a limited colour palette, while the stroke branch parameterises each stroke using Bézier curves to render an image, subsequently evaluated by a high-level recognition module. We quantify the efficiency of visual communication by measuring the recognition accuracy achieved with machine vision. The model then optimises the control points and colour choices for each stroke to maximise recognition accuracy with minimal strokes and colours. Experimental results show that our model achieves superior performance in high-level recognition tasks, delivering artistic expression and aesthetic appeal, especially in abstract sketches. Additionally, our approach shows promise as an efficient bit-level image compression technique, outperforming traditional methods.

Emergence of Painting Ability via Recognition-Driven Evolution

TL;DR

The paper asks whether painting ability can emerge under evolutionary pressures that optimize visual communication. It introduces a two-branch system that generates vector paintings from images: a Stroke Branch that optimizes Bézier strokes and a Palette Branch that learns a restricted colour palette, both guided by a CLIP-based high-level recognition objective and a scenario complexity estimator. The key contributions include emergent painting ability under evolutionary pressure, a dataset and estimator for painting scenario complexity, and demonstrations of high recognition accuracy with few strokes and colours plus potential for low bit rate image compression. The work connects AI driven painting emergence to historical artistic practices and presents practical implications for efficient content generation and compression.

Abstract

From Paleolithic cave paintings to Impressionism, human painting has evolved to depict increasingly complex and detailed scenes, conveying more nuanced messages. This paper attempts to emerge this artistic capability by simulating the evolutionary pressures that enhance visual communication efficiency. Specifically, we present a model with a stroke branch and a palette branch that together simulate human-like painting. The palette branch learns a limited colour palette, while the stroke branch parameterises each stroke using Bézier curves to render an image, subsequently evaluated by a high-level recognition module. We quantify the efficiency of visual communication by measuring the recognition accuracy achieved with machine vision. The model then optimises the control points and colour choices for each stroke to maximise recognition accuracy with minimal strokes and colours. Experimental results show that our model achieves superior performance in high-level recognition tasks, delivering artistic expression and aesthetic appeal, especially in abstract sketches. Additionally, our approach shows promise as an efficient bit-level image compression technique, outperforming traditional methods.
Paper Structure (18 sections, 3 equations, 17 figures, 2 tables)

This paper contains 18 sections, 3 equations, 17 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Painting Generation
  4. Image Compression
  5. Painting Model
  6. Overall Architecture
  7. Stroke Branch
  8. Palette Branch
  9. Recognition Module
  10. Scenario Complexity Estimation
  11. Category Criterion
  12. Composition Criterion
  13. Scenarios Complexity Estimation
  14. Experiments
  15. Quantitative Evaluation
  16. ...and 3 more sections

Figures (17)

  • Figure 1: From cave paintings to Impressionism: the evolution of human painting is reflected in the increasing scenario complexity, enhanced recognition accuracy, and greater abstraction and colour richness.
  • Figure 2: Overview of the painting model. The model generates human-like paintings using a Stroke Branch for stroke optimization and a Palette Branch for colour palette learning. A high-level recognition module is followed to evaluate the recognition accuracy of the generated painting.
  • Figure 3: The process of stroke initialization and the generation of the Reference Colour Palette.
  • Figure 4: Overview of the scenario complexity estimation system.
  • Figure 5: Generated paintings with different numbers of strokes. From top to bottom, both $S_{\text{black}}$ and $S_{\text{colour}}$ is produced using 4, 8, 16, and 32 strokes.
  • ...and 12 more figures