kweimann/poe-learning-layouts

---
license: mit
---
# Learning layouts in Path of Exile with Vision Transformers: A proof of concept

<video controls autoplay src="https://cdn-uploads.huggingface.co/production/uploads/650c55bc9169ea73315b6c22/RJ-rTPWwOFUZlA3ydqhZ2.mp4"></video>

Where's the exit? This question often crosses the minds of both newcomers and seasoned players alike. The key lies in understanding the game's layouts, especially during the campaign when taking a wrong turn can significantly slow you down. Our project aims to solve this challenge through machine learning.

We've developed a proof-of-concept for learning layouts in Path of Exile using Vision Transformers. We trained a Vision Transformer to predict the direction of the exit in the A3 Marketplace, relying solely on a video of the minimap. You can see the model in action in the video above: the red arrow indicates the predicted exit direction, while the green arrow shows the actual direction.

Project page: https://github.com/kweimann/poe-learning-layouts

原始信息汇总

学习《流亡之路》中的布局：使用视觉变换器的概念验证

项目概述

本项目旨在通过机器学习解决《流亡之路》中玩家在游戏布局理解上的挑战，特别是在游戏战役中，错误的转弯会显著减慢玩家速度。我们开发了一个概念验证项目，使用视觉变换器（Vision Transformers）来学习《流亡之路》中的布局。

模型训练

我们训练了一个视觉变换器模型，用于预测A3市场中的出口方向，仅依赖于小地图的视频。在视频中，红色箭头表示预测的出口方向，绿色箭头表示实际的出口方向。

项目页面

项目页面链接：https://github.com/kweimann/poe-learning-layouts