---
language:
- zh
- en
license: mit
tags:
- pytorch
- computer-vision
- cookie-run
- game-ai
- cnn
---

# 🍪 Cookie Run AI Bot Model (跑跑薑餅人 AI 決策模型)

基於 PyTorch 訓練的卷積神經網路 (CNN) 模型，用於即時解析《跑跑薑餅人：烤箱大逃亡》遊戲畫面並做出操作決策。

## ⚙️ 模型資訊 (Model Info)
- **架構**: 3層 Conv2D + MaxPool2D + 全連接分類層 (CookieNet)
- **參數量**: 約 1,152 萬 (11.5M params)
- **輸入尺寸**: 200 x 112 (RGB / BGR 影像，正規化至 0.0 ~ 1.0)
- **輸出動作 (Classes)**:
  - `0: None` (無動作 / 正常奔跑)
  - `1: Jump` (跳躍)
  - `2: Slide` (滑行)
  - `3: Enter` (發動技能)

## 🚀 如何在 Python 中使用 (Usage)

```python
import torch
import cv2
import numpy as np
from model import CookieNet

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = CookieNet().to(device)
model.load_state_dict(torch.load("cookie_bot2.pth", map_location=device))
model.eval()

# 影像前處理 (Resize to 200x112)
img = cv2.imread("screenshot.jpg")
img_resized = cv2.resize(img, (200, 112))
img_tensor = torch.tensor(np.transpose(img_resized.astype(np.float32) / 255.0, (2, 0, 1))).unsqueeze(0).to(device)

with torch.no_grad():
    output = model(img_tensor)
    action = torch.argmax(output, dim=1).item()
    labels = ["None", "Jump", "Slide", "Enter"]
    print(f"預測動作: {labels[action]}")
```