Metin2 (MMORPG) Vision Bot

Automatic Bot for Massive Dungeons Passing based on Windows API, YoloV8 object detection, statistical methods from OpenCV, Tesseract-OCR and spaCy virtualised with Hyper-V (Win11+CUDA)

Demo

demo_dung_stage4_5.mp4

demo_idle_metins.mp4

demo_dung_metins.mp4

Key Packages

• Keyboard and Mouse controll: pynput
• Custom Data Annotation: Roboflow
• Object detection: ultralytics, OpenCV
• Messages handling: Tesseract-OCR, spaCy

Funcionalities

• Autonomic Dungeon Passing
• Autonomic Metin Stones Destroying among all of the channels
• Idle Exp

All of those modes can handle boundary situations such as game crash or logging out.

YoloV8

Model prepared with Ultralytics YOLOv8.0.196 🚀 Python-3.10.12 torch-2.1.0+cu121 CUDA:0 (Tesla T4, 15102MiB)

Class	Images	Instances	Precision*	Recall*	mAP50*	mAP50-95*
all	31	36	0.993	0.939	0.974	0.828
boss_gnoll_cpt	31	11	0.983	0.818	0.931	0.739
metin_polany	31	12	0.996	1	0.995	0.867
npc_straznik	31	13	0.999	1	0.995	0.877

* - Box level

Custom Dataset

Roboflow Dataset Overview (316 images, 800x600)

Augmentation

• Outputs per training example: 3
• Flip: Horizontal
• Blur: Up to 1px

Annotation Heatmap

Direction Recognition

Grayscale Threshold, Contours Recognition and not triangular shape filter