For your specific use-case, a single object detector alone might not give the best possible results for small object detection. Hence, try adding a SAHI component to your object detection pipeline. This sliding window approach for global context sharing, helps a lot. You can also check out our video on the SAHI technique: ua-cam.com/video/UuOjJKxn-M8/v-deo.html
*384*x640 20ms (yolo v8) vs *640*x640 30ms (yolo v9)
- Mmm WHY V9 slower than V8?!
p.s. check resolution 1:52 ^ )
You are correct! v8 infers on smaller image size while v9 infers on larger. The other point on layer and model optimisation is also valid.
I'm working on a project where a small object I'm detecting,so which model I can prefer??
For your specific use-case, a single object detector alone might not give the best possible results for small object detection.
Hence, try adding a SAHI component to your object detection pipeline. This sliding window approach for global context sharing, helps a lot.
You can also check out our video on the SAHI technique: ua-cam.com/video/UuOjJKxn-M8/v-deo.html