2024_GEP

AI model for 2024 Chungbuk Science Exhibition

with: DuvIsaac
If you want to try it, visit this link

AI-Based Grilled Edibility Predictor Model (GEP Model)

This project introduces an AI model that analyzes images to determine the optimal timing to flip meat while grilling.
Using YOLOv5 for object detection, the model classifies cooking stages based on visual changes and temperature.
It helps users avoid overcooking by giving real-time feedback on meat doneness.
Two model structures (GEP1 and GEP2) were tested: full image classification and a hybrid with regression.
Achieved mAP 0.7+, confirming reliable prediction.

Code Overview

graph.ipynb

Notebook for plotting graphs.

• plot_r: Used to plot a single graph.
• Functions above plot_r: Handle multiple plots and additional features.
• Ignore everything below that.

Img_extraction.py

Extracts frames from a video at specific time intervals.

Check the comments in the code

split_folder.py

Splits images with labels into train, test, and valid sets based on a specified ratio.

Check the comments in the code

split_files.py

After splitting into train, test, and valid, this script separates image files and label files into corresponding folders.

Other Scripts

Just read through them or check the filenames — they should be self-explanatory.

Note

There’s no script for generating the data.yaml file — please create it manually.

---

More Information

• labels: Contains all data (images + labels) in one place.
• dataset: Contains the split data (train, test, valid) used for model training.

About 1234 and 123456789 folders

• 1234: Used in the model submitted for the exhibition. Contains only classes 1 to 4.
• 123456789: Contains all classes from 1 to 9.

Class Meaning

Class	Description
1	Already flipped
2	About to flip (1)
3	Flipped (1)
4	About to flip (2)
5	Flipped (2)
6	About to flip (3)
7	Flipped (3)
8	About to flip (4)
9	Flipped (4 - Finished)

---

Excluded Data

• exp02_c3_0055 was excluded due to blurry/shaky footage.