Metadata
- Author: Sushovan Saha
- Full Title: YOLOv8.1 on Custom Dataset — Logo Detection
- URL: https://medium.com/@sushovansaha95/yolov8-1-on-custom-dataset-logo-detection-8915286999ef
Highlights
- Main challenge while training YOLO with custom dataset is preparing the dataset. Here we will be using Flickr27 as our image dataset which contains 27 different brand logos. We have to download and unzip this. Folder contains annotations.txt which consists of -
Press enter or click to view image in full size
(View Highlight) - For Ex — there is an image inside images folder img001.jpg and all its class, x-norm,y-norm,width-norm and height-norm will be stored in space separated manner inside labels folder in img001.txt file. So if we have n images inside images folder then we will also have n .txt files inside labels folder.
Press enter or click to view image in full size
(View Highlight) - Now our dataset is prepared in desired format. Now we split the dataset in train and val set. Now YOLO training module accepts train and val folder path in .yaml file along with that we provide the total no. of classes and class names. structure of the .yaml file - (View Highlight)