Dataset, tracking, small objects #17195
Replies: 2 comments
-
👋 Hello @123louisbe, thank you for your interest in Ultralytics 🚀! This is an automated response, and an Ultralytics engineer will assist you soon. For your project on detecting Asian hornets with YOLOv8, here are a few pointers:
If this is a 🐛 Bug Report, make sure to provide a minimum reproducible example to assist with debugging. For custom training or setup queries, please share detailed information, including dataset samples and training logs, to help facilitate a more in-depth discussion. Upgrade to the latest version of the Join the Ultralytics community on platforms like Discord for real-time discussions, Discourse for detailed topics, or the Subreddit for broader community engagement. You’ll find these spaces helpful for sharing insights and troubleshooting common issues. Happy coding 😃! |
Beta Was this translation helpful? Give feedback.
-
@123louisbe for your dataset, ensure it's well-annotated and consider balancing it further if needed. For tracking, you can use the tracker to extract movement vectors and calculate mean directions. To improve small object detection, try adjusting the model's input size and anchor settings. For more details, visit our documentation at https://docs.ultralytics.com/datasets/track/. |
Beta Was this translation helpful? Give feedback.
-
Hello, i am currently working on a project to detect asian hornets with YOLOv8. The idea is that i place a bait, and film this top down so i can detect the hornets and see in what direction they fly away.
My first question is about the dataset i will be using. The dataset i use consists of about 1500 instances of asian hornets, 300 instances of bees and also 300 of wasps. Only the hornets are annotadet, is this a balanced and good dataset to use?
My second question is about tracking. I understand that Ultralytics implemented a tracker in the YOLOv8 model. Can i extract a vector from this info, make a mean directional vector for all detections and then paste this vector on the video?
Third, i need to be able to identify the hornets, which are 'small objects'. How do i tweak the YOLOv8 model to do this best.
Beta Was this translation helpful? Give feedback.
All reactions