[PYTHON] ASCII generator

• [PYTHON] ASCII generator
  • Fork message
  • Introduction
  • Changes
  • Rationale
    • v1.0 Single-threaded performance
    • v2.0 - Multi-threaded performance
  • Performance optimization
    • v1.1 - use mss instead of pyautogui
    • v3.0 - sliding window + np.mean + threading
    • v3.0.1 - Split screenshot and np + minor optimizations
    • v3.2 - Split mean and text + minor optimizations
  • Next steps

Fork message

可播彩六

Introduction

Original readme could be found at https://github.com/vietnh1009/ASCII-generator.

This repository is a fork of the original repository, featuring stream-to-stream conversion to ASCII art.

Changes

• Implemented stream-to-stream conversion.
• Still supports file-to-file conversion.
• Optimized performance by splitting functions into separate threads.
• Enhanced screen capture speed using mss.
• Replaced nested loops with np.mean for improved efficiency.

Rationale

v1.0 Single-threaded performance

Before threading, the typical fps is 7-8 (4K, stream-to-stream, file-to-file not supported yet)

v2.0 - Multi-threaded performance

Separating major functions into separate threads with a shared queue of size 1 significantly improves performance.

After splitting the functions into separate threads, the typical fps is 9-10 (4K, file-to-file)

The nested loop version (main) takes approximately 0.09 seconds, serving as the program's bottleneck. Note that the screenshot time here is inaccurately measured, as it includes the wait time for putting the image into the queue while the main function continues to process the queue.

Performance optimization

Overall performance (fps) is now determined by the most time-consuming thread, which we will split into separate threads.

v1.1 - use mss instead of pyautogui

The screenshot time is reduced from 0.1 to 0.04 seconds, increasing fps by 5-10 at 4K resolution.

v3.0 - sliding window + np.mean + threading

After replacing the nested loops with np.mean, the typical fps is 17-20 (4K, file-to-file)

v3.0.1 - Split screenshot and np + minor optimizations

The time distribution is now more balanced across each component.

v3.2 - Split mean and text + minor optimizations

Splitting mean and text decreases bottleneck time to 0.06 seconds, yielding a theoretical performance of 16.67 fps and an actual performance of 17-20 fps. (4K, file-to-file)

Next steps

Move time-consuming functions to separate threads to further improve performance.

• Split mean and text will increase little delay but improve performance.