Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evaluating Machine Learning Approaches for ASCII Art Generation

Sai Coumar, Zachary Kingston

TL;DR

This work compares modern deep learning architectures (CNN, ResNet, MobileNetV2, MLP) with classical methods (k-NN, SVM, Random Forest) for generating structure-based ASCII art from images on a synthetic tile dataset. Through SSIM and i2v metrics plus runtime analysis, it shows that Random Forest can match CNN performance with substantially lower compute, while deeper networks can suffer from overmatching on low-dimensional inputs. The study highlights the value of simple, interpretable models for ASCII-art synthesis and provides open-source tooling to support reproducibility and further exploration. Overall, it advances understanding of model selection in low-dimensional image-to-text tasks and suggests domain-tailored hybrid approaches for improved ASCII art generation.

Abstract

Generating structured ASCII art using computational techniques demands a careful interplay between aesthetic representation and computational precision, requiring models that can effectively translate visual information into symbolic text characters. Although Convolutional Neural Networks (CNNs) have shown promise in this domain, the comparative performance of deep learning architectures and classical machine learning methods remains unexplored. This paper explores the application of contemporary ML and DL methods to generate structured ASCII art, focusing on three key criteria: fidelity, character classification accuracy, and output quality. We investigate deep learning architectures, including Multilayer Perceptrons (MLPs), ResNet, and MobileNetV2, alongside classical approaches such as Random Forests, Support Vector Machines (SVMs) and k-Nearest Neighbors (k-NN), trained on an augmented synthetic dataset of ASCII characters. Our results show that complex neural network architectures often fall short in producing high-quality ASCII art, whereas classical machine learning classifiers, despite their simplicity, achieve performance similar to CNNs. Our findings highlight the strength of classical methods in bridging model simplicity with output quality, offering new insights into ASCII art synthesis and machine learning on image data with low dimensionality.

Evaluating Machine Learning Approaches for ASCII Art Generation

TL;DR

This work compares modern deep learning architectures (CNN, ResNet, MobileNetV2, MLP) with classical methods (k-NN, SVM, Random Forest) for generating structure-based ASCII art from images on a synthetic tile dataset. Through SSIM and i2v metrics plus runtime analysis, it shows that Random Forest can match CNN performance with substantially lower compute, while deeper networks can suffer from overmatching on low-dimensional inputs. The study highlights the value of simple, interpretable models for ASCII-art synthesis and provides open-source tooling to support reproducibility and further exploration. Overall, it advances understanding of model selection in low-dimensional image-to-text tasks and suggests domain-tailored hybrid approaches for improved ASCII art generation.

Abstract

Generating structured ASCII art using computational techniques demands a careful interplay between aesthetic representation and computational precision, requiring models that can effectively translate visual information into symbolic text characters. Although Convolutional Neural Networks (CNNs) have shown promise in this domain, the comparative performance of deep learning architectures and classical machine learning methods remains unexplored. This paper explores the application of contemporary ML and DL methods to generate structured ASCII art, focusing on three key criteria: fidelity, character classification accuracy, and output quality. We investigate deep learning architectures, including Multilayer Perceptrons (MLPs), ResNet, and MobileNetV2, alongside classical approaches such as Random Forests, Support Vector Machines (SVMs) and k-Nearest Neighbors (k-NN), trained on an augmented synthetic dataset of ASCII characters. Our results show that complex neural network architectures often fall short in producing high-quality ASCII art, whereas classical machine learning classifiers, despite their simplicity, achieve performance similar to CNNs. Our findings highlight the strength of classical methods in bridging model simplicity with output quality, offering new insights into ASCII art synthesis and machine learning on image data with low dimensionality.

Paper Structure

This paper contains 11 sections, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Image Preprocessing
  5. Tile Conversion
  6. Machine Learning Methods
  7. Materials
  8. Dataset Synthesis
  9. Hardware Platforms
  10. Results
  11. Conclusion

Figures (6)

  • Figure 1: Progression of ASCII Conversion: (a) Original image, (b) Intermediate structural extraction, (c) Final ASCII result (using k-NN).
  • Figure 2: Effects of Tile Size on Image Quality
  • Figure 3: Autoencoder preprocessing decreases structural fidelity and creates overmatching
  • Figure 4: Comparing various techniques for character classification shows that Random Forest matches CNNs in visual output quality
  • Figure 5: Histogram of Gradients does not significantly affect output quality for classical ML models.
  • ...and 1 more figures