Marikhaker/object-detector-telegram-bot

Video object detection using Ultralytics YOLOv8 and Telegram-bot

Run by starting bot.py.

Demo

demo.mp4

Acrhitecture

Training results

Trained YOLOv8s for 22 epoches for 14 classes. Trained on part of Open-Images-v7 dataset: 15k for train, 1k for validate. Dataset downloaded with FiftyOne.

custom_14_cls = ["Clothing", "Tree", "Human face", "Footwear", "Window",
"Building", "Chair", "Fashion accessory",
"Food", "Glasses", "Sunglasses", "Vehicle", "Drink", "Animal"]

Training curves

Confusion matrix

Classification report